]>
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 "qemu/error-report.h" | |
28 | #include "hw/hw.h" | |
29 | #include "hw/pci/msi.h" | |
30 | #include "hw/s390x/adapter.h" | |
31 | #include "exec/gdbstub.h" | |
32 | #include "sysemu/kvm_int.h" | |
33 | #include "qemu/bswap.h" | |
34 | #include "exec/memory.h" | |
35 | #include "exec/ram_addr.h" | |
36 | #include "exec/address-spaces.h" | |
37 | #include "qemu/event_notifier.h" | |
38 | #include "trace.h" | |
39 | #include "hw/irq.h" | |
40 | ||
41 | #include "hw/boards.h" | |
42 | ||
43 | /* This check must be after config-host.h is included */ | |
44 | #ifdef CONFIG_EVENTFD | |
45 | #include <sys/eventfd.h> | |
46 | #endif | |
47 | ||
48 | /* KVM uses PAGE_SIZE in its definition of KVM_COALESCED_MMIO_MAX. We | |
49 | * need to use the real host PAGE_SIZE, as that's what KVM will use. | |
50 | */ | |
51 | #define PAGE_SIZE getpagesize() | |
52 | ||
53 | //#define DEBUG_KVM | |
54 | ||
55 | #ifdef DEBUG_KVM | |
56 | #define DPRINTF(fmt, ...) \ | |
57 | do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0) | |
58 | #else | |
59 | #define DPRINTF(fmt, ...) \ | |
60 | do { } while (0) | |
61 | #endif | |
62 | ||
63 | #define KVM_MSI_HASHTAB_SIZE 256 | |
64 | ||
65 | struct KVMState | |
66 | { | |
67 | AccelState parent_obj; | |
68 | ||
69 | int nr_slots; | |
70 | int fd; | |
71 | int vmfd; | |
72 | int coalesced_mmio; | |
73 | struct kvm_coalesced_mmio_ring *coalesced_mmio_ring; | |
74 | bool coalesced_flush_in_progress; | |
75 | int broken_set_mem_region; | |
76 | int vcpu_events; | |
77 | int robust_singlestep; | |
78 | int debugregs; | |
79 | #ifdef KVM_CAP_SET_GUEST_DEBUG | |
80 | struct kvm_sw_breakpoint_head kvm_sw_breakpoints; | |
81 | #endif | |
82 | int many_ioeventfds; | |
83 | int intx_set_mask; | |
84 | /* The man page (and posix) say ioctl numbers are signed int, but | |
85 | * they're not. Linux, glibc and *BSD all treat ioctl numbers as | |
86 | * unsigned, and treating them as signed here can break things */ | |
87 | unsigned irq_set_ioctl; | |
88 | unsigned int sigmask_len; | |
89 | GHashTable *gsimap; | |
90 | #ifdef KVM_CAP_IRQ_ROUTING | |
91 | struct kvm_irq_routing *irq_routes; | |
92 | int nr_allocated_irq_routes; | |
93 | uint32_t *used_gsi_bitmap; | |
94 | unsigned int gsi_count; | |
95 | QTAILQ_HEAD(msi_hashtab, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE]; | |
96 | #endif | |
97 | KVMMemoryListener memory_listener; | |
98 | }; | |
99 | ||
100 | KVMState *kvm_state; | |
101 | bool kvm_kernel_irqchip; | |
102 | bool kvm_async_interrupts_allowed; | |
103 | bool kvm_halt_in_kernel_allowed; | |
104 | bool kvm_eventfds_allowed; | |
105 | bool kvm_irqfds_allowed; | |
106 | bool kvm_resamplefds_allowed; | |
107 | bool kvm_msi_via_irqfd_allowed; | |
108 | bool kvm_gsi_routing_allowed; | |
109 | bool kvm_gsi_direct_mapping; | |
110 | bool kvm_allowed; | |
111 | bool kvm_readonly_mem_allowed; | |
112 | bool kvm_vm_attributes_allowed; | |
113 | bool kvm_direct_msi_allowed; | |
114 | bool kvm_ioeventfd_any_length_allowed; | |
115 | ||
116 | static const KVMCapabilityInfo kvm_required_capabilites[] = { | |
117 | KVM_CAP_INFO(USER_MEMORY), | |
118 | KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS), | |
119 | KVM_CAP_LAST_INFO | |
120 | }; | |
121 | ||
122 | static KVMSlot *kvm_get_free_slot(KVMMemoryListener *kml) | |
123 | { | |
124 | KVMState *s = kvm_state; | |
125 | int i; | |
126 | ||
127 | for (i = 0; i < s->nr_slots; i++) { | |
128 | if (kml->slots[i].memory_size == 0) { | |
129 | return &kml->slots[i]; | |
130 | } | |
131 | } | |
132 | ||
133 | return NULL; | |
134 | } | |
135 | ||
136 | bool kvm_has_free_slot(MachineState *ms) | |
137 | { | |
138 | KVMState *s = KVM_STATE(ms->accelerator); | |
139 | ||
140 | return kvm_get_free_slot(&s->memory_listener); | |
141 | } | |
142 | ||
143 | static KVMSlot *kvm_alloc_slot(KVMMemoryListener *kml) | |
144 | { | |
145 | KVMSlot *slot = kvm_get_free_slot(kml); | |
146 | ||
147 | if (slot) { | |
148 | return slot; | |
149 | } | |
150 | ||
151 | fprintf(stderr, "%s: no free slot available\n", __func__); | |
152 | abort(); | |
153 | } | |
154 | ||
155 | static KVMSlot *kvm_lookup_matching_slot(KVMMemoryListener *kml, | |
156 | hwaddr start_addr, | |
157 | hwaddr end_addr) | |
158 | { | |
159 | KVMState *s = kvm_state; | |
160 | int i; | |
161 | ||
162 | for (i = 0; i < s->nr_slots; i++) { | |
163 | KVMSlot *mem = &kml->slots[i]; | |
164 | ||
165 | if (start_addr == mem->start_addr && | |
166 | end_addr == mem->start_addr + mem->memory_size) { | |
167 | return mem; | |
168 | } | |
169 | } | |
170 | ||
171 | return NULL; | |
172 | } | |
173 | ||
174 | /* | |
175 | * Find overlapping slot with lowest start address | |
176 | */ | |
177 | static KVMSlot *kvm_lookup_overlapping_slot(KVMMemoryListener *kml, | |
178 | hwaddr start_addr, | |
179 | hwaddr end_addr) | |
180 | { | |
181 | KVMState *s = kvm_state; | |
182 | KVMSlot *found = NULL; | |
183 | int i; | |
184 | ||
185 | for (i = 0; i < s->nr_slots; i++) { | |
186 | KVMSlot *mem = &kml->slots[i]; | |
187 | ||
188 | if (mem->memory_size == 0 || | |
189 | (found && found->start_addr < mem->start_addr)) { | |
190 | continue; | |
191 | } | |
192 | ||
193 | if (end_addr > mem->start_addr && | |
194 | start_addr < mem->start_addr + mem->memory_size) { | |
195 | found = mem; | |
196 | } | |
197 | } | |
198 | ||
199 | return found; | |
200 | } | |
201 | ||
202 | int kvm_physical_memory_addr_from_host(KVMState *s, void *ram, | |
203 | hwaddr *phys_addr) | |
204 | { | |
205 | KVMMemoryListener *kml = &s->memory_listener; | |
206 | int i; | |
207 | ||
208 | for (i = 0; i < s->nr_slots; i++) { | |
209 | KVMSlot *mem = &kml->slots[i]; | |
210 | ||
211 | if (ram >= mem->ram && ram < mem->ram + mem->memory_size) { | |
212 | *phys_addr = mem->start_addr + (ram - mem->ram); | |
213 | return 1; | |
214 | } | |
215 | } | |
216 | ||
217 | return 0; | |
218 | } | |
219 | ||
220 | static int kvm_set_user_memory_region(KVMMemoryListener *kml, KVMSlot *slot) | |
221 | { | |
222 | KVMState *s = kvm_state; | |
223 | struct kvm_userspace_memory_region mem; | |
224 | ||
225 | mem.slot = slot->slot | (kml->as_id << 16); | |
226 | mem.guest_phys_addr = slot->start_addr; | |
227 | mem.userspace_addr = (unsigned long)slot->ram; | |
228 | mem.flags = slot->flags; | |
229 | ||
230 | if (slot->memory_size && mem.flags & KVM_MEM_READONLY) { | |
231 | /* Set the slot size to 0 before setting the slot to the desired | |
232 | * value. This is needed based on KVM commit 75d61fbc. */ | |
233 | mem.memory_size = 0; | |
234 | kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); | |
235 | } | |
236 | mem.memory_size = slot->memory_size; | |
237 | return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem); | |
238 | } | |
239 | ||
240 | int kvm_init_vcpu(CPUState *cpu) | |
241 | { | |
242 | KVMState *s = kvm_state; | |
243 | long mmap_size; | |
244 | int ret; | |
245 | ||
246 | DPRINTF("kvm_init_vcpu\n"); | |
247 | ||
248 | ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)kvm_arch_vcpu_id(cpu)); | |
249 | if (ret < 0) { | |
250 | DPRINTF("kvm_create_vcpu failed\n"); | |
251 | goto err; | |
252 | } | |
253 | ||
254 | cpu->kvm_fd = ret; | |
255 | cpu->kvm_state = s; | |
256 | cpu->kvm_vcpu_dirty = true; | |
257 | ||
258 | mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0); | |
259 | if (mmap_size < 0) { | |
260 | ret = mmap_size; | |
261 | DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n"); | |
262 | goto err; | |
263 | } | |
264 | ||
265 | cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, | |
266 | cpu->kvm_fd, 0); | |
267 | if (cpu->kvm_run == MAP_FAILED) { | |
268 | ret = -errno; | |
269 | DPRINTF("mmap'ing vcpu state failed\n"); | |
270 | goto err; | |
271 | } | |
272 | ||
273 | if (s->coalesced_mmio && !s->coalesced_mmio_ring) { | |
274 | s->coalesced_mmio_ring = | |
275 | (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE; | |
276 | } | |
277 | ||
278 | ret = kvm_arch_init_vcpu(cpu); | |
279 | err: | |
280 | return ret; | |
281 | } | |
282 | ||
283 | /* | |
284 | * dirty pages logging control | |
285 | */ | |
286 | ||
287 | static int kvm_mem_flags(MemoryRegion *mr) | |
288 | { | |
289 | bool readonly = mr->readonly || memory_region_is_romd(mr); | |
290 | int flags = 0; | |
291 | ||
292 | if (memory_region_get_dirty_log_mask(mr) != 0) { | |
293 | flags |= KVM_MEM_LOG_DIRTY_PAGES; | |
294 | } | |
295 | if (readonly && kvm_readonly_mem_allowed) { | |
296 | flags |= KVM_MEM_READONLY; | |
297 | } | |
298 | return flags; | |
299 | } | |
300 | ||
301 | static int kvm_slot_update_flags(KVMMemoryListener *kml, KVMSlot *mem, | |
302 | MemoryRegion *mr) | |
303 | { | |
304 | int old_flags; | |
305 | ||
306 | old_flags = mem->flags; | |
307 | mem->flags = kvm_mem_flags(mr); | |
308 | ||
309 | /* If nothing changed effectively, no need to issue ioctl */ | |
310 | if (mem->flags == old_flags) { | |
311 | return 0; | |
312 | } | |
313 | ||
314 | return kvm_set_user_memory_region(kml, mem); | |
315 | } | |
316 | ||
317 | static int kvm_section_update_flags(KVMMemoryListener *kml, | |
318 | MemoryRegionSection *section) | |
319 | { | |
320 | hwaddr phys_addr = section->offset_within_address_space; | |
321 | ram_addr_t size = int128_get64(section->size); | |
322 | KVMSlot *mem = kvm_lookup_matching_slot(kml, phys_addr, phys_addr + size); | |
323 | ||
324 | if (mem == NULL) { | |
325 | return 0; | |
326 | } else { | |
327 | return kvm_slot_update_flags(kml, mem, section->mr); | |
328 | } | |
329 | } | |
330 | ||
331 | static void kvm_log_start(MemoryListener *listener, | |
332 | MemoryRegionSection *section, | |
333 | int old, int new) | |
334 | { | |
335 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); | |
336 | int r; | |
337 | ||
338 | if (old != 0) { | |
339 | return; | |
340 | } | |
341 | ||
342 | r = kvm_section_update_flags(kml, section); | |
343 | if (r < 0) { | |
344 | abort(); | |
345 | } | |
346 | } | |
347 | ||
348 | static void kvm_log_stop(MemoryListener *listener, | |
349 | MemoryRegionSection *section, | |
350 | int old, int new) | |
351 | { | |
352 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); | |
353 | int r; | |
354 | ||
355 | if (new != 0) { | |
356 | return; | |
357 | } | |
358 | ||
359 | r = kvm_section_update_flags(kml, section); | |
360 | if (r < 0) { | |
361 | abort(); | |
362 | } | |
363 | } | |
364 | ||
365 | /* get kvm's dirty pages bitmap and update qemu's */ | |
366 | static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section, | |
367 | unsigned long *bitmap) | |
368 | { | |
369 | ram_addr_t start = section->offset_within_region + section->mr->ram_addr; | |
370 | ram_addr_t pages = int128_get64(section->size) / getpagesize(); | |
371 | ||
372 | cpu_physical_memory_set_dirty_lebitmap(bitmap, start, pages); | |
373 | return 0; | |
374 | } | |
375 | ||
376 | #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1)) | |
377 | ||
378 | /** | |
379 | * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space | |
380 | * This function updates qemu's dirty bitmap using | |
381 | * memory_region_set_dirty(). This means all bits are set | |
382 | * to dirty. | |
383 | * | |
384 | * @start_add: start of logged region. | |
385 | * @end_addr: end of logged region. | |
386 | */ | |
387 | static int kvm_physical_sync_dirty_bitmap(KVMMemoryListener *kml, | |
388 | MemoryRegionSection *section) | |
389 | { | |
390 | KVMState *s = kvm_state; | |
391 | unsigned long size, allocated_size = 0; | |
392 | struct kvm_dirty_log d = {}; | |
393 | KVMSlot *mem; | |
394 | int ret = 0; | |
395 | hwaddr start_addr = section->offset_within_address_space; | |
396 | hwaddr end_addr = start_addr + int128_get64(section->size); | |
397 | ||
398 | d.dirty_bitmap = NULL; | |
399 | while (start_addr < end_addr) { | |
400 | mem = kvm_lookup_overlapping_slot(kml, start_addr, end_addr); | |
401 | if (mem == NULL) { | |
402 | break; | |
403 | } | |
404 | ||
405 | /* XXX bad kernel interface alert | |
406 | * For dirty bitmap, kernel allocates array of size aligned to | |
407 | * bits-per-long. But for case when the kernel is 64bits and | |
408 | * the userspace is 32bits, userspace can't align to the same | |
409 | * bits-per-long, since sizeof(long) is different between kernel | |
410 | * and user space. This way, userspace will provide buffer which | |
411 | * may be 4 bytes less than the kernel will use, resulting in | |
412 | * userspace memory corruption (which is not detectable by valgrind | |
413 | * too, in most cases). | |
414 | * So for now, let's align to 64 instead of HOST_LONG_BITS here, in | |
415 | * a hope that sizeof(long) wont become >8 any time soon. | |
416 | */ | |
417 | size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS), | |
418 | /*HOST_LONG_BITS*/ 64) / 8; | |
419 | if (!d.dirty_bitmap) { | |
420 | d.dirty_bitmap = g_malloc(size); | |
421 | } else if (size > allocated_size) { | |
422 | d.dirty_bitmap = g_realloc(d.dirty_bitmap, size); | |
423 | } | |
424 | allocated_size = size; | |
425 | memset(d.dirty_bitmap, 0, allocated_size); | |
426 | ||
427 | d.slot = mem->slot | (kml->as_id << 16); | |
428 | if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) { | |
429 | DPRINTF("ioctl failed %d\n", errno); | |
430 | ret = -1; | |
431 | break; | |
432 | } | |
433 | ||
434 | kvm_get_dirty_pages_log_range(section, d.dirty_bitmap); | |
435 | start_addr = mem->start_addr + mem->memory_size; | |
436 | } | |
437 | g_free(d.dirty_bitmap); | |
438 | ||
439 | return ret; | |
440 | } | |
441 | ||
442 | static void kvm_coalesce_mmio_region(MemoryListener *listener, | |
443 | MemoryRegionSection *secion, | |
444 | hwaddr start, hwaddr size) | |
445 | { | |
446 | KVMState *s = kvm_state; | |
447 | ||
448 | if (s->coalesced_mmio) { | |
449 | struct kvm_coalesced_mmio_zone zone; | |
450 | ||
451 | zone.addr = start; | |
452 | zone.size = size; | |
453 | zone.pad = 0; | |
454 | ||
455 | (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone); | |
456 | } | |
457 | } | |
458 | ||
459 | static void kvm_uncoalesce_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_UNREGISTER_COALESCED_MMIO, &zone); | |
473 | } | |
474 | } | |
475 | ||
476 | int kvm_check_extension(KVMState *s, unsigned int extension) | |
477 | { | |
478 | int ret; | |
479 | ||
480 | ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
481 | if (ret < 0) { | |
482 | ret = 0; | |
483 | } | |
484 | ||
485 | return ret; | |
486 | } | |
487 | ||
488 | int kvm_vm_check_extension(KVMState *s, unsigned int extension) | |
489 | { | |
490 | int ret; | |
491 | ||
492 | ret = kvm_vm_ioctl(s, KVM_CHECK_EXTENSION, extension); | |
493 | if (ret < 0) { | |
494 | /* VM wide version not implemented, use global one instead */ | |
495 | ret = kvm_check_extension(s, extension); | |
496 | } | |
497 | ||
498 | return ret; | |
499 | } | |
500 | ||
501 | static uint32_t adjust_ioeventfd_endianness(uint32_t val, uint32_t size) | |
502 | { | |
503 | #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN) | |
504 | /* The kernel expects ioeventfd values in HOST_WORDS_BIGENDIAN | |
505 | * endianness, but the memory core hands them in target endianness. | |
506 | * For example, PPC is always treated as big-endian even if running | |
507 | * on KVM and on PPC64LE. Correct here. | |
508 | */ | |
509 | switch (size) { | |
510 | case 2: | |
511 | val = bswap16(val); | |
512 | break; | |
513 | case 4: | |
514 | val = bswap32(val); | |
515 | break; | |
516 | } | |
517 | #endif | |
518 | return val; | |
519 | } | |
520 | ||
521 | static int kvm_set_ioeventfd_mmio(int fd, hwaddr addr, uint32_t val, | |
522 | bool assign, uint32_t size, bool datamatch) | |
523 | { | |
524 | int ret; | |
525 | struct kvm_ioeventfd iofd = { | |
526 | .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0, | |
527 | .addr = addr, | |
528 | .len = size, | |
529 | .flags = 0, | |
530 | .fd = fd, | |
531 | }; | |
532 | ||
533 | if (!kvm_enabled()) { | |
534 | return -ENOSYS; | |
535 | } | |
536 | ||
537 | if (datamatch) { | |
538 | iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH; | |
539 | } | |
540 | if (!assign) { | |
541 | iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
542 | } | |
543 | ||
544 | ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd); | |
545 | ||
546 | if (ret < 0) { | |
547 | return -errno; | |
548 | } | |
549 | ||
550 | return 0; | |
551 | } | |
552 | ||
553 | static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val, | |
554 | bool assign, uint32_t size, bool datamatch) | |
555 | { | |
556 | struct kvm_ioeventfd kick = { | |
557 | .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0, | |
558 | .addr = addr, | |
559 | .flags = KVM_IOEVENTFD_FLAG_PIO, | |
560 | .len = size, | |
561 | .fd = fd, | |
562 | }; | |
563 | int r; | |
564 | if (!kvm_enabled()) { | |
565 | return -ENOSYS; | |
566 | } | |
567 | if (datamatch) { | |
568 | kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH; | |
569 | } | |
570 | if (!assign) { | |
571 | kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN; | |
572 | } | |
573 | r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick); | |
574 | if (r < 0) { | |
575 | return r; | |
576 | } | |
577 | return 0; | |
578 | } | |
579 | ||
580 | ||
581 | static int kvm_check_many_ioeventfds(void) | |
582 | { | |
583 | /* Userspace can use ioeventfd for io notification. This requires a host | |
584 | * that supports eventfd(2) and an I/O thread; since eventfd does not | |
585 | * support SIGIO it cannot interrupt the vcpu. | |
586 | * | |
587 | * Older kernels have a 6 device limit on the KVM io bus. Find out so we | |
588 | * can avoid creating too many ioeventfds. | |
589 | */ | |
590 | #if defined(CONFIG_EVENTFD) | |
591 | int ioeventfds[7]; | |
592 | int i, ret = 0; | |
593 | for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) { | |
594 | ioeventfds[i] = eventfd(0, EFD_CLOEXEC); | |
595 | if (ioeventfds[i] < 0) { | |
596 | break; | |
597 | } | |
598 | ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true); | |
599 | if (ret < 0) { | |
600 | close(ioeventfds[i]); | |
601 | break; | |
602 | } | |
603 | } | |
604 | ||
605 | /* Decide whether many devices are supported or not */ | |
606 | ret = i == ARRAY_SIZE(ioeventfds); | |
607 | ||
608 | while (i-- > 0) { | |
609 | kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true); | |
610 | close(ioeventfds[i]); | |
611 | } | |
612 | return ret; | |
613 | #else | |
614 | return 0; | |
615 | #endif | |
616 | } | |
617 | ||
618 | static const KVMCapabilityInfo * | |
619 | kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list) | |
620 | { | |
621 | while (list->name) { | |
622 | if (!kvm_check_extension(s, list->value)) { | |
623 | return list; | |
624 | } | |
625 | list++; | |
626 | } | |
627 | return NULL; | |
628 | } | |
629 | ||
630 | static void kvm_set_phys_mem(KVMMemoryListener *kml, | |
631 | MemoryRegionSection *section, bool add) | |
632 | { | |
633 | KVMState *s = kvm_state; | |
634 | KVMSlot *mem, old; | |
635 | int err; | |
636 | MemoryRegion *mr = section->mr; | |
637 | bool writeable = !mr->readonly && !mr->rom_device; | |
638 | hwaddr start_addr = section->offset_within_address_space; | |
639 | ram_addr_t size = int128_get64(section->size); | |
640 | void *ram = NULL; | |
641 | unsigned delta; | |
642 | ||
643 | /* kvm works in page size chunks, but the function may be called | |
644 | with sub-page size and unaligned start address. Pad the start | |
645 | address to next and truncate size to previous page boundary. */ | |
646 | delta = qemu_real_host_page_size - (start_addr & ~qemu_real_host_page_mask); | |
647 | delta &= ~qemu_real_host_page_mask; | |
648 | if (delta > size) { | |
649 | return; | |
650 | } | |
651 | start_addr += delta; | |
652 | size -= delta; | |
653 | size &= qemu_real_host_page_mask; | |
654 | if (!size || (start_addr & ~qemu_real_host_page_mask)) { | |
655 | return; | |
656 | } | |
657 | ||
658 | if (!memory_region_is_ram(mr)) { | |
659 | if (writeable || !kvm_readonly_mem_allowed) { | |
660 | return; | |
661 | } else if (!mr->romd_mode) { | |
662 | /* If the memory device is not in romd_mode, then we actually want | |
663 | * to remove the kvm memory slot so all accesses will trap. */ | |
664 | add = false; | |
665 | } | |
666 | } | |
667 | ||
668 | ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta; | |
669 | ||
670 | while (1) { | |
671 | mem = kvm_lookup_overlapping_slot(kml, start_addr, start_addr + size); | |
672 | if (!mem) { | |
673 | break; | |
674 | } | |
675 | ||
676 | if (add && start_addr >= mem->start_addr && | |
677 | (start_addr + size <= mem->start_addr + mem->memory_size) && | |
678 | (ram - start_addr == mem->ram - mem->start_addr)) { | |
679 | /* The new slot fits into the existing one and comes with | |
680 | * identical parameters - update flags and done. */ | |
681 | kvm_slot_update_flags(kml, mem, mr); | |
682 | return; | |
683 | } | |
684 | ||
685 | old = *mem; | |
686 | ||
687 | if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) { | |
688 | kvm_physical_sync_dirty_bitmap(kml, section); | |
689 | } | |
690 | ||
691 | /* unregister the overlapping slot */ | |
692 | mem->memory_size = 0; | |
693 | err = kvm_set_user_memory_region(kml, mem); | |
694 | if (err) { | |
695 | fprintf(stderr, "%s: error unregistering overlapping slot: %s\n", | |
696 | __func__, strerror(-err)); | |
697 | abort(); | |
698 | } | |
699 | ||
700 | /* Workaround for older KVM versions: we can't join slots, even not by | |
701 | * unregistering the previous ones and then registering the larger | |
702 | * slot. We have to maintain the existing fragmentation. Sigh. | |
703 | * | |
704 | * This workaround assumes that the new slot starts at the same | |
705 | * address as the first existing one. If not or if some overlapping | |
706 | * slot comes around later, we will fail (not seen in practice so far) | |
707 | * - and actually require a recent KVM version. */ | |
708 | if (s->broken_set_mem_region && | |
709 | old.start_addr == start_addr && old.memory_size < size && add) { | |
710 | mem = kvm_alloc_slot(kml); | |
711 | mem->memory_size = old.memory_size; | |
712 | mem->start_addr = old.start_addr; | |
713 | mem->ram = old.ram; | |
714 | mem->flags = kvm_mem_flags(mr); | |
715 | ||
716 | err = kvm_set_user_memory_region(kml, mem); | |
717 | if (err) { | |
718 | fprintf(stderr, "%s: error updating slot: %s\n", __func__, | |
719 | strerror(-err)); | |
720 | abort(); | |
721 | } | |
722 | ||
723 | start_addr += old.memory_size; | |
724 | ram += old.memory_size; | |
725 | size -= old.memory_size; | |
726 | continue; | |
727 | } | |
728 | ||
729 | /* register prefix slot */ | |
730 | if (old.start_addr < start_addr) { | |
731 | mem = kvm_alloc_slot(kml); | |
732 | mem->memory_size = start_addr - old.start_addr; | |
733 | mem->start_addr = old.start_addr; | |
734 | mem->ram = old.ram; | |
735 | mem->flags = kvm_mem_flags(mr); | |
736 | ||
737 | err = kvm_set_user_memory_region(kml, mem); | |
738 | if (err) { | |
739 | fprintf(stderr, "%s: error registering prefix slot: %s\n", | |
740 | __func__, strerror(-err)); | |
741 | #ifdef TARGET_PPC | |
742 | fprintf(stderr, "%s: This is probably because your kernel's " \ | |
743 | "PAGE_SIZE is too big. Please try to use 4k " \ | |
744 | "PAGE_SIZE!\n", __func__); | |
745 | #endif | |
746 | abort(); | |
747 | } | |
748 | } | |
749 | ||
750 | /* register suffix slot */ | |
751 | if (old.start_addr + old.memory_size > start_addr + size) { | |
752 | ram_addr_t size_delta; | |
753 | ||
754 | mem = kvm_alloc_slot(kml); | |
755 | mem->start_addr = start_addr + size; | |
756 | size_delta = mem->start_addr - old.start_addr; | |
757 | mem->memory_size = old.memory_size - size_delta; | |
758 | mem->ram = old.ram + size_delta; | |
759 | mem->flags = kvm_mem_flags(mr); | |
760 | ||
761 | err = kvm_set_user_memory_region(kml, mem); | |
762 | if (err) { | |
763 | fprintf(stderr, "%s: error registering suffix slot: %s\n", | |
764 | __func__, strerror(-err)); | |
765 | abort(); | |
766 | } | |
767 | } | |
768 | } | |
769 | ||
770 | /* in case the KVM bug workaround already "consumed" the new slot */ | |
771 | if (!size) { | |
772 | return; | |
773 | } | |
774 | if (!add) { | |
775 | return; | |
776 | } | |
777 | mem = kvm_alloc_slot(kml); | |
778 | mem->memory_size = size; | |
779 | mem->start_addr = start_addr; | |
780 | mem->ram = ram; | |
781 | mem->flags = kvm_mem_flags(mr); | |
782 | ||
783 | err = kvm_set_user_memory_region(kml, mem); | |
784 | if (err) { | |
785 | fprintf(stderr, "%s: error registering slot: %s\n", __func__, | |
786 | strerror(-err)); | |
787 | abort(); | |
788 | } | |
789 | } | |
790 | ||
791 | static void kvm_region_add(MemoryListener *listener, | |
792 | MemoryRegionSection *section) | |
793 | { | |
794 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); | |
795 | ||
796 | memory_region_ref(section->mr); | |
797 | kvm_set_phys_mem(kml, section, true); | |
798 | } | |
799 | ||
800 | static void kvm_region_del(MemoryListener *listener, | |
801 | MemoryRegionSection *section) | |
802 | { | |
803 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); | |
804 | ||
805 | kvm_set_phys_mem(kml, section, false); | |
806 | memory_region_unref(section->mr); | |
807 | } | |
808 | ||
809 | static void kvm_log_sync(MemoryListener *listener, | |
810 | MemoryRegionSection *section) | |
811 | { | |
812 | KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener); | |
813 | int r; | |
814 | ||
815 | r = kvm_physical_sync_dirty_bitmap(kml, section); | |
816 | if (r < 0) { | |
817 | abort(); | |
818 | } | |
819 | } | |
820 | ||
821 | static void kvm_mem_ioeventfd_add(MemoryListener *listener, | |
822 | MemoryRegionSection *section, | |
823 | bool match_data, uint64_t data, | |
824 | EventNotifier *e) | |
825 | { | |
826 | int fd = event_notifier_get_fd(e); | |
827 | int r; | |
828 | ||
829 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, | |
830 | data, true, int128_get64(section->size), | |
831 | match_data); | |
832 | if (r < 0) { | |
833 | fprintf(stderr, "%s: error adding ioeventfd: %s\n", | |
834 | __func__, strerror(-r)); | |
835 | abort(); | |
836 | } | |
837 | } | |
838 | ||
839 | static void kvm_mem_ioeventfd_del(MemoryListener *listener, | |
840 | MemoryRegionSection *section, | |
841 | bool match_data, uint64_t data, | |
842 | EventNotifier *e) | |
843 | { | |
844 | int fd = event_notifier_get_fd(e); | |
845 | int r; | |
846 | ||
847 | r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space, | |
848 | data, false, int128_get64(section->size), | |
849 | match_data); | |
850 | if (r < 0) { | |
851 | abort(); | |
852 | } | |
853 | } | |
854 | ||
855 | static void kvm_io_ioeventfd_add(MemoryListener *listener, | |
856 | MemoryRegionSection *section, | |
857 | bool match_data, uint64_t data, | |
858 | EventNotifier *e) | |
859 | { | |
860 | int fd = event_notifier_get_fd(e); | |
861 | int r; | |
862 | ||
863 | r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space, | |
864 | data, true, int128_get64(section->size), | |
865 | match_data); | |
866 | if (r < 0) { | |
867 | fprintf(stderr, "%s: error adding ioeventfd: %s\n", | |
868 | __func__, strerror(-r)); | |
869 | abort(); | |
870 | } | |
871 | } | |
872 | ||
873 | static void kvm_io_ioeventfd_del(MemoryListener *listener, | |
874 | MemoryRegionSection *section, | |
875 | bool match_data, uint64_t data, | |
876 | EventNotifier *e) | |
877 | ||
878 | { | |
879 | int fd = event_notifier_get_fd(e); | |
880 | int r; | |
881 | ||
882 | r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space, | |
883 | data, false, int128_get64(section->size), | |
884 | match_data); | |
885 | if (r < 0) { | |
886 | abort(); | |
887 | } | |
888 | } | |
889 | ||
890 | void kvm_memory_listener_register(KVMState *s, KVMMemoryListener *kml, | |
891 | AddressSpace *as, int as_id) | |
892 | { | |
893 | int i; | |
894 | ||
895 | kml->slots = g_malloc0(s->nr_slots * sizeof(KVMSlot)); | |
896 | kml->as_id = as_id; | |
897 | ||
898 | for (i = 0; i < s->nr_slots; i++) { | |
899 | kml->slots[i].slot = i; | |
900 | } | |
901 | ||
902 | kml->listener.region_add = kvm_region_add; | |
903 | kml->listener.region_del = kvm_region_del; | |
904 | kml->listener.log_start = kvm_log_start; | |
905 | kml->listener.log_stop = kvm_log_stop; | |
906 | kml->listener.log_sync = kvm_log_sync; | |
907 | kml->listener.priority = 10; | |
908 | ||
909 | memory_listener_register(&kml->listener, as); | |
910 | } | |
911 | ||
912 | static MemoryListener kvm_io_listener = { | |
913 | .eventfd_add = kvm_io_ioeventfd_add, | |
914 | .eventfd_del = kvm_io_ioeventfd_del, | |
915 | .priority = 10, | |
916 | }; | |
917 | ||
918 | static void kvm_handle_interrupt(CPUState *cpu, int mask) | |
919 | { | |
920 | cpu->interrupt_request |= mask; | |
921 | ||
922 | if (!qemu_cpu_is_self(cpu)) { | |
923 | qemu_cpu_kick(cpu); | |
924 | } | |
925 | } | |
926 | ||
927 | int kvm_set_irq(KVMState *s, int irq, int level) | |
928 | { | |
929 | struct kvm_irq_level event; | |
930 | int ret; | |
931 | ||
932 | assert(kvm_async_interrupts_enabled()); | |
933 | ||
934 | event.level = level; | |
935 | event.irq = irq; | |
936 | ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event); | |
937 | if (ret < 0) { | |
938 | perror("kvm_set_irq"); | |
939 | abort(); | |
940 | } | |
941 | ||
942 | return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status; | |
943 | } | |
944 | ||
945 | #ifdef KVM_CAP_IRQ_ROUTING | |
946 | typedef struct KVMMSIRoute { | |
947 | struct kvm_irq_routing_entry kroute; | |
948 | QTAILQ_ENTRY(KVMMSIRoute) entry; | |
949 | } KVMMSIRoute; | |
950 | ||
951 | static void set_gsi(KVMState *s, unsigned int gsi) | |
952 | { | |
953 | s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32); | |
954 | } | |
955 | ||
956 | static void clear_gsi(KVMState *s, unsigned int gsi) | |
957 | { | |
958 | s->used_gsi_bitmap[gsi / 32] &= ~(1U << (gsi % 32)); | |
959 | } | |
960 | ||
961 | void kvm_init_irq_routing(KVMState *s) | |
962 | { | |
963 | int gsi_count, i; | |
964 | ||
965 | gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING) - 1; | |
966 | if (gsi_count > 0) { | |
967 | unsigned int gsi_bits, i; | |
968 | ||
969 | /* Round up so we can search ints using ffs */ | |
970 | gsi_bits = ALIGN(gsi_count, 32); | |
971 | s->used_gsi_bitmap = g_malloc0(gsi_bits / 8); | |
972 | s->gsi_count = gsi_count; | |
973 | ||
974 | /* Mark any over-allocated bits as already in use */ | |
975 | for (i = gsi_count; i < gsi_bits; i++) { | |
976 | set_gsi(s, i); | |
977 | } | |
978 | } | |
979 | ||
980 | s->irq_routes = g_malloc0(sizeof(*s->irq_routes)); | |
981 | s->nr_allocated_irq_routes = 0; | |
982 | ||
983 | if (!kvm_direct_msi_allowed) { | |
984 | for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) { | |
985 | QTAILQ_INIT(&s->msi_hashtab[i]); | |
986 | } | |
987 | } | |
988 | ||
989 | kvm_arch_init_irq_routing(s); | |
990 | } | |
991 | ||
992 | void kvm_irqchip_commit_routes(KVMState *s) | |
993 | { | |
994 | int ret; | |
995 | ||
996 | s->irq_routes->flags = 0; | |
997 | ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes); | |
998 | assert(ret == 0); | |
999 | } | |
1000 | ||
1001 | static void kvm_add_routing_entry(KVMState *s, | |
1002 | struct kvm_irq_routing_entry *entry) | |
1003 | { | |
1004 | struct kvm_irq_routing_entry *new; | |
1005 | int n, size; | |
1006 | ||
1007 | if (s->irq_routes->nr == s->nr_allocated_irq_routes) { | |
1008 | n = s->nr_allocated_irq_routes * 2; | |
1009 | if (n < 64) { | |
1010 | n = 64; | |
1011 | } | |
1012 | size = sizeof(struct kvm_irq_routing); | |
1013 | size += n * sizeof(*new); | |
1014 | s->irq_routes = g_realloc(s->irq_routes, size); | |
1015 | s->nr_allocated_irq_routes = n; | |
1016 | } | |
1017 | n = s->irq_routes->nr++; | |
1018 | new = &s->irq_routes->entries[n]; | |
1019 | ||
1020 | *new = *entry; | |
1021 | ||
1022 | set_gsi(s, entry->gsi); | |
1023 | } | |
1024 | ||
1025 | static int kvm_update_routing_entry(KVMState *s, | |
1026 | struct kvm_irq_routing_entry *new_entry) | |
1027 | { | |
1028 | struct kvm_irq_routing_entry *entry; | |
1029 | int n; | |
1030 | ||
1031 | for (n = 0; n < s->irq_routes->nr; n++) { | |
1032 | entry = &s->irq_routes->entries[n]; | |
1033 | if (entry->gsi != new_entry->gsi) { | |
1034 | continue; | |
1035 | } | |
1036 | ||
1037 | if(!memcmp(entry, new_entry, sizeof *entry)) { | |
1038 | return 0; | |
1039 | } | |
1040 | ||
1041 | *entry = *new_entry; | |
1042 | ||
1043 | kvm_irqchip_commit_routes(s); | |
1044 | ||
1045 | return 0; | |
1046 | } | |
1047 | ||
1048 | return -ESRCH; | |
1049 | } | |
1050 | ||
1051 | void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin) | |
1052 | { | |
1053 | struct kvm_irq_routing_entry e = {}; | |
1054 | ||
1055 | assert(pin < s->gsi_count); | |
1056 | ||
1057 | e.gsi = irq; | |
1058 | e.type = KVM_IRQ_ROUTING_IRQCHIP; | |
1059 | e.flags = 0; | |
1060 | e.u.irqchip.irqchip = irqchip; | |
1061 | e.u.irqchip.pin = pin; | |
1062 | kvm_add_routing_entry(s, &e); | |
1063 | } | |
1064 | ||
1065 | void kvm_irqchip_release_virq(KVMState *s, int virq) | |
1066 | { | |
1067 | struct kvm_irq_routing_entry *e; | |
1068 | int i; | |
1069 | ||
1070 | if (kvm_gsi_direct_mapping()) { | |
1071 | return; | |
1072 | } | |
1073 | ||
1074 | for (i = 0; i < s->irq_routes->nr; i++) { | |
1075 | e = &s->irq_routes->entries[i]; | |
1076 | if (e->gsi == virq) { | |
1077 | s->irq_routes->nr--; | |
1078 | *e = s->irq_routes->entries[s->irq_routes->nr]; | |
1079 | } | |
1080 | } | |
1081 | clear_gsi(s, virq); | |
1082 | } | |
1083 | ||
1084 | static unsigned int kvm_hash_msi(uint32_t data) | |
1085 | { | |
1086 | /* This is optimized for IA32 MSI layout. However, no other arch shall | |
1087 | * repeat the mistake of not providing a direct MSI injection API. */ | |
1088 | return data & 0xff; | |
1089 | } | |
1090 | ||
1091 | static void kvm_flush_dynamic_msi_routes(KVMState *s) | |
1092 | { | |
1093 | KVMMSIRoute *route, *next; | |
1094 | unsigned int hash; | |
1095 | ||
1096 | for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) { | |
1097 | QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) { | |
1098 | kvm_irqchip_release_virq(s, route->kroute.gsi); | |
1099 | QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry); | |
1100 | g_free(route); | |
1101 | } | |
1102 | } | |
1103 | } | |
1104 | ||
1105 | static int kvm_irqchip_get_virq(KVMState *s) | |
1106 | { | |
1107 | uint32_t *word = s->used_gsi_bitmap; | |
1108 | int max_words = ALIGN(s->gsi_count, 32) / 32; | |
1109 | int i, zeroes; | |
1110 | ||
1111 | /* | |
1112 | * PIC and IOAPIC share the first 16 GSI numbers, thus the available | |
1113 | * GSI numbers are more than the number of IRQ route. Allocating a GSI | |
1114 | * number can succeed even though a new route entry cannot be added. | |
1115 | * When this happens, flush dynamic MSI entries to free IRQ route entries. | |
1116 | */ | |
1117 | if (!kvm_direct_msi_allowed && s->irq_routes->nr == s->gsi_count) { | |
1118 | kvm_flush_dynamic_msi_routes(s); | |
1119 | } | |
1120 | ||
1121 | /* Return the lowest unused GSI in the bitmap */ | |
1122 | for (i = 0; i < max_words; i++) { | |
1123 | zeroes = ctz32(~word[i]); | |
1124 | if (zeroes == 32) { | |
1125 | continue; | |
1126 | } | |
1127 | ||
1128 | return zeroes + i * 32; | |
1129 | } | |
1130 | return -ENOSPC; | |
1131 | ||
1132 | } | |
1133 | ||
1134 | static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg) | |
1135 | { | |
1136 | unsigned int hash = kvm_hash_msi(msg.data); | |
1137 | KVMMSIRoute *route; | |
1138 | ||
1139 | QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) { | |
1140 | if (route->kroute.u.msi.address_lo == (uint32_t)msg.address && | |
1141 | route->kroute.u.msi.address_hi == (msg.address >> 32) && | |
1142 | route->kroute.u.msi.data == le32_to_cpu(msg.data)) { | |
1143 | return route; | |
1144 | } | |
1145 | } | |
1146 | return NULL; | |
1147 | } | |
1148 | ||
1149 | int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg) | |
1150 | { | |
1151 | struct kvm_msi msi; | |
1152 | KVMMSIRoute *route; | |
1153 | ||
1154 | if (kvm_direct_msi_allowed) { | |
1155 | msi.address_lo = (uint32_t)msg.address; | |
1156 | msi.address_hi = msg.address >> 32; | |
1157 | msi.data = le32_to_cpu(msg.data); | |
1158 | msi.flags = 0; | |
1159 | memset(msi.pad, 0, sizeof(msi.pad)); | |
1160 | ||
1161 | return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi); | |
1162 | } | |
1163 | ||
1164 | route = kvm_lookup_msi_route(s, msg); | |
1165 | if (!route) { | |
1166 | int virq; | |
1167 | ||
1168 | virq = kvm_irqchip_get_virq(s); | |
1169 | if (virq < 0) { | |
1170 | return virq; | |
1171 | } | |
1172 | ||
1173 | route = g_malloc0(sizeof(KVMMSIRoute)); | |
1174 | route->kroute.gsi = virq; | |
1175 | route->kroute.type = KVM_IRQ_ROUTING_MSI; | |
1176 | route->kroute.flags = 0; | |
1177 | route->kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1178 | route->kroute.u.msi.address_hi = msg.address >> 32; | |
1179 | route->kroute.u.msi.data = le32_to_cpu(msg.data); | |
1180 | ||
1181 | kvm_add_routing_entry(s, &route->kroute); | |
1182 | kvm_irqchip_commit_routes(s); | |
1183 | ||
1184 | QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route, | |
1185 | entry); | |
1186 | } | |
1187 | ||
1188 | assert(route->kroute.type == KVM_IRQ_ROUTING_MSI); | |
1189 | ||
1190 | return kvm_set_irq(s, route->kroute.gsi, 1); | |
1191 | } | |
1192 | ||
1193 | int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg, PCIDevice *dev) | |
1194 | { | |
1195 | struct kvm_irq_routing_entry kroute = {}; | |
1196 | int virq; | |
1197 | ||
1198 | if (kvm_gsi_direct_mapping()) { | |
1199 | return kvm_arch_msi_data_to_gsi(msg.data); | |
1200 | } | |
1201 | ||
1202 | if (!kvm_gsi_routing_enabled()) { | |
1203 | return -ENOSYS; | |
1204 | } | |
1205 | ||
1206 | virq = kvm_irqchip_get_virq(s); | |
1207 | if (virq < 0) { | |
1208 | return virq; | |
1209 | } | |
1210 | ||
1211 | kroute.gsi = virq; | |
1212 | kroute.type = KVM_IRQ_ROUTING_MSI; | |
1213 | kroute.flags = 0; | |
1214 | kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1215 | kroute.u.msi.address_hi = msg.address >> 32; | |
1216 | kroute.u.msi.data = le32_to_cpu(msg.data); | |
1217 | if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data, dev)) { | |
1218 | kvm_irqchip_release_virq(s, virq); | |
1219 | return -EINVAL; | |
1220 | } | |
1221 | ||
1222 | kvm_add_routing_entry(s, &kroute); | |
1223 | kvm_irqchip_commit_routes(s); | |
1224 | ||
1225 | return virq; | |
1226 | } | |
1227 | ||
1228 | int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg, | |
1229 | PCIDevice *dev) | |
1230 | { | |
1231 | struct kvm_irq_routing_entry kroute = {}; | |
1232 | ||
1233 | if (kvm_gsi_direct_mapping()) { | |
1234 | return 0; | |
1235 | } | |
1236 | ||
1237 | if (!kvm_irqchip_in_kernel()) { | |
1238 | return -ENOSYS; | |
1239 | } | |
1240 | ||
1241 | kroute.gsi = virq; | |
1242 | kroute.type = KVM_IRQ_ROUTING_MSI; | |
1243 | kroute.flags = 0; | |
1244 | kroute.u.msi.address_lo = (uint32_t)msg.address; | |
1245 | kroute.u.msi.address_hi = msg.address >> 32; | |
1246 | kroute.u.msi.data = le32_to_cpu(msg.data); | |
1247 | if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data, dev)) { | |
1248 | return -EINVAL; | |
1249 | } | |
1250 | ||
1251 | return kvm_update_routing_entry(s, &kroute); | |
1252 | } | |
1253 | ||
1254 | static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int rfd, int virq, | |
1255 | bool assign) | |
1256 | { | |
1257 | struct kvm_irqfd irqfd = { | |
1258 | .fd = fd, | |
1259 | .gsi = virq, | |
1260 | .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN, | |
1261 | }; | |
1262 | ||
1263 | if (rfd != -1) { | |
1264 | irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE; | |
1265 | irqfd.resamplefd = rfd; | |
1266 | } | |
1267 | ||
1268 | if (!kvm_irqfds_enabled()) { | |
1269 | return -ENOSYS; | |
1270 | } | |
1271 | ||
1272 | return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd); | |
1273 | } | |
1274 | ||
1275 | int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter) | |
1276 | { | |
1277 | struct kvm_irq_routing_entry kroute = {}; | |
1278 | int virq; | |
1279 | ||
1280 | if (!kvm_gsi_routing_enabled()) { | |
1281 | return -ENOSYS; | |
1282 | } | |
1283 | ||
1284 | virq = kvm_irqchip_get_virq(s); | |
1285 | if (virq < 0) { | |
1286 | return virq; | |
1287 | } | |
1288 | ||
1289 | kroute.gsi = virq; | |
1290 | kroute.type = KVM_IRQ_ROUTING_S390_ADAPTER; | |
1291 | kroute.flags = 0; | |
1292 | kroute.u.adapter.summary_addr = adapter->summary_addr; | |
1293 | kroute.u.adapter.ind_addr = adapter->ind_addr; | |
1294 | kroute.u.adapter.summary_offset = adapter->summary_offset; | |
1295 | kroute.u.adapter.ind_offset = adapter->ind_offset; | |
1296 | kroute.u.adapter.adapter_id = adapter->adapter_id; | |
1297 | ||
1298 | kvm_add_routing_entry(s, &kroute); | |
1299 | ||
1300 | return virq; | |
1301 | } | |
1302 | ||
1303 | #else /* !KVM_CAP_IRQ_ROUTING */ | |
1304 | ||
1305 | void kvm_init_irq_routing(KVMState *s) | |
1306 | { | |
1307 | } | |
1308 | ||
1309 | void kvm_irqchip_release_virq(KVMState *s, int virq) | |
1310 | { | |
1311 | } | |
1312 | ||
1313 | int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg) | |
1314 | { | |
1315 | abort(); | |
1316 | } | |
1317 | ||
1318 | int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg) | |
1319 | { | |
1320 | return -ENOSYS; | |
1321 | } | |
1322 | ||
1323 | int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter) | |
1324 | { | |
1325 | return -ENOSYS; | |
1326 | } | |
1327 | ||
1328 | static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign) | |
1329 | { | |
1330 | abort(); | |
1331 | } | |
1332 | ||
1333 | int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg) | |
1334 | { | |
1335 | return -ENOSYS; | |
1336 | } | |
1337 | #endif /* !KVM_CAP_IRQ_ROUTING */ | |
1338 | ||
1339 | int kvm_irqchip_add_irqfd_notifier_gsi(KVMState *s, EventNotifier *n, | |
1340 | EventNotifier *rn, int virq) | |
1341 | { | |
1342 | return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), | |
1343 | rn ? event_notifier_get_fd(rn) : -1, virq, true); | |
1344 | } | |
1345 | ||
1346 | int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState *s, EventNotifier *n, | |
1347 | int virq) | |
1348 | { | |
1349 | return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), -1, virq, | |
1350 | false); | |
1351 | } | |
1352 | ||
1353 | int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n, | |
1354 | EventNotifier *rn, qemu_irq irq) | |
1355 | { | |
1356 | gpointer key, gsi; | |
1357 | gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi); | |
1358 | ||
1359 | if (!found) { | |
1360 | return -ENXIO; | |
1361 | } | |
1362 | return kvm_irqchip_add_irqfd_notifier_gsi(s, n, rn, GPOINTER_TO_INT(gsi)); | |
1363 | } | |
1364 | ||
1365 | int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, | |
1366 | qemu_irq irq) | |
1367 | { | |
1368 | gpointer key, gsi; | |
1369 | gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi); | |
1370 | ||
1371 | if (!found) { | |
1372 | return -ENXIO; | |
1373 | } | |
1374 | return kvm_irqchip_remove_irqfd_notifier_gsi(s, n, GPOINTER_TO_INT(gsi)); | |
1375 | } | |
1376 | ||
1377 | void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi) | |
1378 | { | |
1379 | g_hash_table_insert(s->gsimap, irq, GINT_TO_POINTER(gsi)); | |
1380 | } | |
1381 | ||
1382 | static void kvm_irqchip_create(MachineState *machine, KVMState *s) | |
1383 | { | |
1384 | int ret; | |
1385 | ||
1386 | if (kvm_check_extension(s, KVM_CAP_IRQCHIP)) { | |
1387 | ; | |
1388 | } else if (kvm_check_extension(s, KVM_CAP_S390_IRQCHIP)) { | |
1389 | ret = kvm_vm_enable_cap(s, KVM_CAP_S390_IRQCHIP, 0); | |
1390 | if (ret < 0) { | |
1391 | fprintf(stderr, "Enable kernel irqchip failed: %s\n", strerror(-ret)); | |
1392 | exit(1); | |
1393 | } | |
1394 | } else { | |
1395 | return; | |
1396 | } | |
1397 | ||
1398 | /* First probe and see if there's a arch-specific hook to create the | |
1399 | * in-kernel irqchip for us */ | |
1400 | ret = kvm_arch_irqchip_create(s); | |
1401 | if (ret == 0) { | |
1402 | ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP); | |
1403 | } | |
1404 | if (ret < 0) { | |
1405 | fprintf(stderr, "Create kernel irqchip failed: %s\n", strerror(-ret)); | |
1406 | exit(1); | |
1407 | } | |
1408 | ||
1409 | kvm_kernel_irqchip = true; | |
1410 | /* If we have an in-kernel IRQ chip then we must have asynchronous | |
1411 | * interrupt delivery (though the reverse is not necessarily true) | |
1412 | */ | |
1413 | kvm_async_interrupts_allowed = true; | |
1414 | kvm_halt_in_kernel_allowed = true; | |
1415 | ||
1416 | kvm_init_irq_routing(s); | |
1417 | ||
1418 | s->gsimap = g_hash_table_new(g_direct_hash, g_direct_equal); | |
1419 | } | |
1420 | ||
1421 | /* Find number of supported CPUs using the recommended | |
1422 | * procedure from the kernel API documentation to cope with | |
1423 | * older kernels that may be missing capabilities. | |
1424 | */ | |
1425 | static int kvm_recommended_vcpus(KVMState *s) | |
1426 | { | |
1427 | int ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS); | |
1428 | return (ret) ? ret : 4; | |
1429 | } | |
1430 | ||
1431 | static int kvm_max_vcpus(KVMState *s) | |
1432 | { | |
1433 | int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS); | |
1434 | return (ret) ? ret : kvm_recommended_vcpus(s); | |
1435 | } | |
1436 | ||
1437 | static int kvm_init(MachineState *ms) | |
1438 | { | |
1439 | MachineClass *mc = MACHINE_GET_CLASS(ms); | |
1440 | static const char upgrade_note[] = | |
1441 | "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n" | |
1442 | "(see http://sourceforge.net/projects/kvm).\n"; | |
1443 | struct { | |
1444 | const char *name; | |
1445 | int num; | |
1446 | } num_cpus[] = { | |
1447 | { "SMP", smp_cpus }, | |
1448 | { "hotpluggable", max_cpus }, | |
1449 | { NULL, } | |
1450 | }, *nc = num_cpus; | |
1451 | int soft_vcpus_limit, hard_vcpus_limit; | |
1452 | KVMState *s; | |
1453 | const KVMCapabilityInfo *missing_cap; | |
1454 | int ret; | |
1455 | int type = 0; | |
1456 | const char *kvm_type; | |
1457 | ||
1458 | s = KVM_STATE(ms->accelerator); | |
1459 | ||
1460 | /* | |
1461 | * On systems where the kernel can support different base page | |
1462 | * sizes, host page size may be different from TARGET_PAGE_SIZE, | |
1463 | * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum | |
1464 | * page size for the system though. | |
1465 | */ | |
1466 | assert(TARGET_PAGE_SIZE <= getpagesize()); | |
1467 | ||
1468 | s->sigmask_len = 8; | |
1469 | ||
1470 | #ifdef KVM_CAP_SET_GUEST_DEBUG | |
1471 | QTAILQ_INIT(&s->kvm_sw_breakpoints); | |
1472 | #endif | |
1473 | s->vmfd = -1; | |
1474 | s->fd = qemu_open("/dev/kvm", O_RDWR); | |
1475 | if (s->fd == -1) { | |
1476 | fprintf(stderr, "Could not access KVM kernel module: %m\n"); | |
1477 | ret = -errno; | |
1478 | goto err; | |
1479 | } | |
1480 | ||
1481 | ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0); | |
1482 | if (ret < KVM_API_VERSION) { | |
1483 | if (ret >= 0) { | |
1484 | ret = -EINVAL; | |
1485 | } | |
1486 | fprintf(stderr, "kvm version too old\n"); | |
1487 | goto err; | |
1488 | } | |
1489 | ||
1490 | if (ret > KVM_API_VERSION) { | |
1491 | ret = -EINVAL; | |
1492 | fprintf(stderr, "kvm version not supported\n"); | |
1493 | goto err; | |
1494 | } | |
1495 | ||
1496 | s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS); | |
1497 | ||
1498 | /* If unspecified, use the default value */ | |
1499 | if (!s->nr_slots) { | |
1500 | s->nr_slots = 32; | |
1501 | } | |
1502 | ||
1503 | /* check the vcpu limits */ | |
1504 | soft_vcpus_limit = kvm_recommended_vcpus(s); | |
1505 | hard_vcpus_limit = kvm_max_vcpus(s); | |
1506 | ||
1507 | while (nc->name) { | |
1508 | if (nc->num > soft_vcpus_limit) { | |
1509 | fprintf(stderr, | |
1510 | "Warning: Number of %s cpus requested (%d) exceeds " | |
1511 | "the recommended cpus supported by KVM (%d)\n", | |
1512 | nc->name, nc->num, soft_vcpus_limit); | |
1513 | ||
1514 | if (nc->num > hard_vcpus_limit) { | |
1515 | fprintf(stderr, "Number of %s cpus requested (%d) exceeds " | |
1516 | "the maximum cpus supported by KVM (%d)\n", | |
1517 | nc->name, nc->num, hard_vcpus_limit); | |
1518 | exit(1); | |
1519 | } | |
1520 | } | |
1521 | nc++; | |
1522 | } | |
1523 | ||
1524 | kvm_type = qemu_opt_get(qemu_get_machine_opts(), "kvm-type"); | |
1525 | if (mc->kvm_type) { | |
1526 | type = mc->kvm_type(kvm_type); | |
1527 | } else if (kvm_type) { | |
1528 | ret = -EINVAL; | |
1529 | fprintf(stderr, "Invalid argument kvm-type=%s\n", kvm_type); | |
1530 | goto err; | |
1531 | } | |
1532 | ||
1533 | do { | |
1534 | ret = kvm_ioctl(s, KVM_CREATE_VM, type); | |
1535 | } while (ret == -EINTR); | |
1536 | ||
1537 | if (ret < 0) { | |
1538 | fprintf(stderr, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret, | |
1539 | strerror(-ret)); | |
1540 | ||
1541 | #ifdef TARGET_S390X | |
1542 | if (ret == -EINVAL) { | |
1543 | fprintf(stderr, | |
1544 | "Host kernel setup problem detected. Please verify:\n"); | |
1545 | fprintf(stderr, "- for kernels supporting the switch_amode or" | |
1546 | " user_mode parameters, whether\n"); | |
1547 | fprintf(stderr, | |
1548 | " user space is running in primary address space\n"); | |
1549 | fprintf(stderr, | |
1550 | "- for kernels supporting the vm.allocate_pgste sysctl, " | |
1551 | "whether it is enabled\n"); | |
1552 | } | |
1553 | #endif | |
1554 | goto err; | |
1555 | } | |
1556 | ||
1557 | s->vmfd = ret; | |
1558 | missing_cap = kvm_check_extension_list(s, kvm_required_capabilites); | |
1559 | if (!missing_cap) { | |
1560 | missing_cap = | |
1561 | kvm_check_extension_list(s, kvm_arch_required_capabilities); | |
1562 | } | |
1563 | if (missing_cap) { | |
1564 | ret = -EINVAL; | |
1565 | fprintf(stderr, "kvm does not support %s\n%s", | |
1566 | missing_cap->name, upgrade_note); | |
1567 | goto err; | |
1568 | } | |
1569 | ||
1570 | s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO); | |
1571 | ||
1572 | s->broken_set_mem_region = 1; | |
1573 | ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS); | |
1574 | if (ret > 0) { | |
1575 | s->broken_set_mem_region = 0; | |
1576 | } | |
1577 | ||
1578 | #ifdef KVM_CAP_VCPU_EVENTS | |
1579 | s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS); | |
1580 | #endif | |
1581 | ||
1582 | s->robust_singlestep = | |
1583 | kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP); | |
1584 | ||
1585 | #ifdef KVM_CAP_DEBUGREGS | |
1586 | s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS); | |
1587 | #endif | |
1588 | ||
1589 | #ifdef KVM_CAP_IRQ_ROUTING | |
1590 | kvm_direct_msi_allowed = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0); | |
1591 | #endif | |
1592 | ||
1593 | s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3); | |
1594 | ||
1595 | s->irq_set_ioctl = KVM_IRQ_LINE; | |
1596 | if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) { | |
1597 | s->irq_set_ioctl = KVM_IRQ_LINE_STATUS; | |
1598 | } | |
1599 | ||
1600 | #ifdef KVM_CAP_READONLY_MEM | |
1601 | kvm_readonly_mem_allowed = | |
1602 | (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0); | |
1603 | #endif | |
1604 | ||
1605 | kvm_eventfds_allowed = | |
1606 | (kvm_check_extension(s, KVM_CAP_IOEVENTFD) > 0); | |
1607 | ||
1608 | kvm_irqfds_allowed = | |
1609 | (kvm_check_extension(s, KVM_CAP_IRQFD) > 0); | |
1610 | ||
1611 | kvm_resamplefds_allowed = | |
1612 | (kvm_check_extension(s, KVM_CAP_IRQFD_RESAMPLE) > 0); | |
1613 | ||
1614 | kvm_vm_attributes_allowed = | |
1615 | (kvm_check_extension(s, KVM_CAP_VM_ATTRIBUTES) > 0); | |
1616 | ||
1617 | kvm_ioeventfd_any_length_allowed = | |
1618 | (kvm_check_extension(s, KVM_CAP_IOEVENTFD_ANY_LENGTH) > 0); | |
1619 | ||
1620 | ret = kvm_arch_init(ms, s); | |
1621 | if (ret < 0) { | |
1622 | goto err; | |
1623 | } | |
1624 | ||
1625 | if (machine_kernel_irqchip_allowed(ms)) { | |
1626 | kvm_irqchip_create(ms, s); | |
1627 | } | |
1628 | ||
1629 | kvm_state = s; | |
1630 | ||
1631 | s->memory_listener.listener.eventfd_add = kvm_mem_ioeventfd_add; | |
1632 | s->memory_listener.listener.eventfd_del = kvm_mem_ioeventfd_del; | |
1633 | s->memory_listener.listener.coalesced_mmio_add = kvm_coalesce_mmio_region; | |
1634 | s->memory_listener.listener.coalesced_mmio_del = kvm_uncoalesce_mmio_region; | |
1635 | ||
1636 | kvm_memory_listener_register(s, &s->memory_listener, | |
1637 | &address_space_memory, 0); | |
1638 | memory_listener_register(&kvm_io_listener, | |
1639 | &address_space_io); | |
1640 | ||
1641 | s->many_ioeventfds = kvm_check_many_ioeventfds(); | |
1642 | ||
1643 | cpu_interrupt_handler = kvm_handle_interrupt; | |
1644 | ||
1645 | return 0; | |
1646 | ||
1647 | err: | |
1648 | assert(ret < 0); | |
1649 | if (s->vmfd >= 0) { | |
1650 | close(s->vmfd); | |
1651 | } | |
1652 | if (s->fd != -1) { | |
1653 | close(s->fd); | |
1654 | } | |
1655 | g_free(s->memory_listener.slots); | |
1656 | ||
1657 | return ret; | |
1658 | } | |
1659 | ||
1660 | void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len) | |
1661 | { | |
1662 | s->sigmask_len = sigmask_len; | |
1663 | } | |
1664 | ||
1665 | static void kvm_handle_io(uint16_t port, MemTxAttrs attrs, void *data, int direction, | |
1666 | int size, uint32_t count) | |
1667 | { | |
1668 | int i; | |
1669 | uint8_t *ptr = data; | |
1670 | ||
1671 | for (i = 0; i < count; i++) { | |
1672 | address_space_rw(&address_space_io, port, attrs, | |
1673 | ptr, size, | |
1674 | direction == KVM_EXIT_IO_OUT); | |
1675 | ptr += size; | |
1676 | } | |
1677 | } | |
1678 | ||
1679 | static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run) | |
1680 | { | |
1681 | fprintf(stderr, "KVM internal error. Suberror: %d\n", | |
1682 | run->internal.suberror); | |
1683 | ||
1684 | if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) { | |
1685 | int i; | |
1686 | ||
1687 | for (i = 0; i < run->internal.ndata; ++i) { | |
1688 | fprintf(stderr, "extra data[%d]: %"PRIx64"\n", | |
1689 | i, (uint64_t)run->internal.data[i]); | |
1690 | } | |
1691 | } | |
1692 | if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) { | |
1693 | fprintf(stderr, "emulation failure\n"); | |
1694 | if (!kvm_arch_stop_on_emulation_error(cpu)) { | |
1695 | cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE); | |
1696 | return EXCP_INTERRUPT; | |
1697 | } | |
1698 | } | |
1699 | /* FIXME: Should trigger a qmp message to let management know | |
1700 | * something went wrong. | |
1701 | */ | |
1702 | return -1; | |
1703 | } | |
1704 | ||
1705 | void kvm_flush_coalesced_mmio_buffer(void) | |
1706 | { | |
1707 | KVMState *s = kvm_state; | |
1708 | ||
1709 | if (s->coalesced_flush_in_progress) { | |
1710 | return; | |
1711 | } | |
1712 | ||
1713 | s->coalesced_flush_in_progress = true; | |
1714 | ||
1715 | if (s->coalesced_mmio_ring) { | |
1716 | struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring; | |
1717 | while (ring->first != ring->last) { | |
1718 | struct kvm_coalesced_mmio *ent; | |
1719 | ||
1720 | ent = &ring->coalesced_mmio[ring->first]; | |
1721 | ||
1722 | cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len); | |
1723 | smp_wmb(); | |
1724 | ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX; | |
1725 | } | |
1726 | } | |
1727 | ||
1728 | s->coalesced_flush_in_progress = false; | |
1729 | } | |
1730 | ||
1731 | static void do_kvm_cpu_synchronize_state(void *arg) | |
1732 | { | |
1733 | CPUState *cpu = arg; | |
1734 | ||
1735 | if (!cpu->kvm_vcpu_dirty) { | |
1736 | kvm_arch_get_registers(cpu); | |
1737 | cpu->kvm_vcpu_dirty = true; | |
1738 | } | |
1739 | } | |
1740 | ||
1741 | void kvm_cpu_synchronize_state(CPUState *cpu) | |
1742 | { | |
1743 | if (!cpu->kvm_vcpu_dirty) { | |
1744 | run_on_cpu(cpu, do_kvm_cpu_synchronize_state, cpu); | |
1745 | } | |
1746 | } | |
1747 | ||
1748 | static void do_kvm_cpu_synchronize_post_reset(void *arg) | |
1749 | { | |
1750 | CPUState *cpu = arg; | |
1751 | ||
1752 | kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE); | |
1753 | cpu->kvm_vcpu_dirty = false; | |
1754 | } | |
1755 | ||
1756 | void kvm_cpu_synchronize_post_reset(CPUState *cpu) | |
1757 | { | |
1758 | run_on_cpu(cpu, do_kvm_cpu_synchronize_post_reset, cpu); | |
1759 | } | |
1760 | ||
1761 | static void do_kvm_cpu_synchronize_post_init(void *arg) | |
1762 | { | |
1763 | CPUState *cpu = arg; | |
1764 | ||
1765 | kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE); | |
1766 | cpu->kvm_vcpu_dirty = false; | |
1767 | } | |
1768 | ||
1769 | void kvm_cpu_synchronize_post_init(CPUState *cpu) | |
1770 | { | |
1771 | run_on_cpu(cpu, do_kvm_cpu_synchronize_post_init, cpu); | |
1772 | } | |
1773 | ||
1774 | int kvm_cpu_exec(CPUState *cpu) | |
1775 | { | |
1776 | struct kvm_run *run = cpu->kvm_run; | |
1777 | int ret, run_ret; | |
1778 | ||
1779 | DPRINTF("kvm_cpu_exec()\n"); | |
1780 | ||
1781 | if (kvm_arch_process_async_events(cpu)) { | |
1782 | cpu->exit_request = 0; | |
1783 | return EXCP_HLT; | |
1784 | } | |
1785 | ||
1786 | qemu_mutex_unlock_iothread(); | |
1787 | ||
1788 | do { | |
1789 | MemTxAttrs attrs; | |
1790 | ||
1791 | if (cpu->kvm_vcpu_dirty) { | |
1792 | kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE); | |
1793 | cpu->kvm_vcpu_dirty = false; | |
1794 | } | |
1795 | ||
1796 | kvm_arch_pre_run(cpu, run); | |
1797 | if (cpu->exit_request) { | |
1798 | DPRINTF("interrupt exit requested\n"); | |
1799 | /* | |
1800 | * KVM requires us to reenter the kernel after IO exits to complete | |
1801 | * instruction emulation. This self-signal will ensure that we | |
1802 | * leave ASAP again. | |
1803 | */ | |
1804 | qemu_cpu_kick_self(); | |
1805 | } | |
1806 | ||
1807 | run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0); | |
1808 | ||
1809 | attrs = kvm_arch_post_run(cpu, run); | |
1810 | ||
1811 | if (run_ret < 0) { | |
1812 | if (run_ret == -EINTR || run_ret == -EAGAIN) { | |
1813 | DPRINTF("io window exit\n"); | |
1814 | ret = EXCP_INTERRUPT; | |
1815 | break; | |
1816 | } | |
1817 | fprintf(stderr, "error: kvm run failed %s\n", | |
1818 | strerror(-run_ret)); | |
1819 | #ifdef TARGET_PPC | |
1820 | if (run_ret == -EBUSY) { | |
1821 | fprintf(stderr, | |
1822 | "This is probably because your SMT is enabled.\n" | |
1823 | "VCPU can only run on primary threads with all " | |
1824 | "secondary threads offline.\n"); | |
1825 | } | |
1826 | #endif | |
1827 | ret = -1; | |
1828 | break; | |
1829 | } | |
1830 | ||
1831 | trace_kvm_run_exit(cpu->cpu_index, run->exit_reason); | |
1832 | switch (run->exit_reason) { | |
1833 | case KVM_EXIT_IO: | |
1834 | DPRINTF("handle_io\n"); | |
1835 | /* Called outside BQL */ | |
1836 | kvm_handle_io(run->io.port, attrs, | |
1837 | (uint8_t *)run + run->io.data_offset, | |
1838 | run->io.direction, | |
1839 | run->io.size, | |
1840 | run->io.count); | |
1841 | ret = 0; | |
1842 | break; | |
1843 | case KVM_EXIT_MMIO: | |
1844 | DPRINTF("handle_mmio\n"); | |
1845 | /* Called outside BQL */ | |
1846 | address_space_rw(&address_space_memory, | |
1847 | run->mmio.phys_addr, attrs, | |
1848 | run->mmio.data, | |
1849 | run->mmio.len, | |
1850 | run->mmio.is_write); | |
1851 | ret = 0; | |
1852 | break; | |
1853 | case KVM_EXIT_IRQ_WINDOW_OPEN: | |
1854 | DPRINTF("irq_window_open\n"); | |
1855 | ret = EXCP_INTERRUPT; | |
1856 | break; | |
1857 | case KVM_EXIT_SHUTDOWN: | |
1858 | DPRINTF("shutdown\n"); | |
1859 | qemu_system_reset_request(); | |
1860 | ret = EXCP_INTERRUPT; | |
1861 | break; | |
1862 | case KVM_EXIT_UNKNOWN: | |
1863 | fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n", | |
1864 | (uint64_t)run->hw.hardware_exit_reason); | |
1865 | ret = -1; | |
1866 | break; | |
1867 | case KVM_EXIT_INTERNAL_ERROR: | |
1868 | ret = kvm_handle_internal_error(cpu, run); | |
1869 | break; | |
1870 | case KVM_EXIT_SYSTEM_EVENT: | |
1871 | switch (run->system_event.type) { | |
1872 | case KVM_SYSTEM_EVENT_SHUTDOWN: | |
1873 | qemu_system_shutdown_request(); | |
1874 | ret = EXCP_INTERRUPT; | |
1875 | break; | |
1876 | case KVM_SYSTEM_EVENT_RESET: | |
1877 | qemu_system_reset_request(); | |
1878 | ret = EXCP_INTERRUPT; | |
1879 | break; | |
1880 | case KVM_SYSTEM_EVENT_CRASH: | |
1881 | qemu_mutex_lock_iothread(); | |
1882 | qemu_system_guest_panicked(); | |
1883 | qemu_mutex_unlock_iothread(); | |
1884 | ret = 0; | |
1885 | break; | |
1886 | default: | |
1887 | DPRINTF("kvm_arch_handle_exit\n"); | |
1888 | ret = kvm_arch_handle_exit(cpu, run); | |
1889 | break; | |
1890 | } | |
1891 | break; | |
1892 | default: | |
1893 | DPRINTF("kvm_arch_handle_exit\n"); | |
1894 | ret = kvm_arch_handle_exit(cpu, run); | |
1895 | break; | |
1896 | } | |
1897 | } while (ret == 0); | |
1898 | ||
1899 | qemu_mutex_lock_iothread(); | |
1900 | ||
1901 | if (ret < 0) { | |
1902 | cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE); | |
1903 | vm_stop(RUN_STATE_INTERNAL_ERROR); | |
1904 | } | |
1905 | ||
1906 | cpu->exit_request = 0; | |
1907 | return ret; | |
1908 | } | |
1909 | ||
1910 | int kvm_ioctl(KVMState *s, int type, ...) | |
1911 | { | |
1912 | int ret; | |
1913 | void *arg; | |
1914 | va_list ap; | |
1915 | ||
1916 | va_start(ap, type); | |
1917 | arg = va_arg(ap, void *); | |
1918 | va_end(ap); | |
1919 | ||
1920 | trace_kvm_ioctl(type, arg); | |
1921 | ret = ioctl(s->fd, type, arg); | |
1922 | if (ret == -1) { | |
1923 | ret = -errno; | |
1924 | } | |
1925 | return ret; | |
1926 | } | |
1927 | ||
1928 | int kvm_vm_ioctl(KVMState *s, int type, ...) | |
1929 | { | |
1930 | int ret; | |
1931 | void *arg; | |
1932 | va_list ap; | |
1933 | ||
1934 | va_start(ap, type); | |
1935 | arg = va_arg(ap, void *); | |
1936 | va_end(ap); | |
1937 | ||
1938 | trace_kvm_vm_ioctl(type, arg); | |
1939 | ret = ioctl(s->vmfd, type, arg); | |
1940 | if (ret == -1) { | |
1941 | ret = -errno; | |
1942 | } | |
1943 | return ret; | |
1944 | } | |
1945 | ||
1946 | int kvm_vcpu_ioctl(CPUState *cpu, int type, ...) | |
1947 | { | |
1948 | int ret; | |
1949 | void *arg; | |
1950 | va_list ap; | |
1951 | ||
1952 | va_start(ap, type); | |
1953 | arg = va_arg(ap, void *); | |
1954 | va_end(ap); | |
1955 | ||
1956 | trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg); | |
1957 | ret = ioctl(cpu->kvm_fd, type, arg); | |
1958 | if (ret == -1) { | |
1959 | ret = -errno; | |
1960 | } | |
1961 | return ret; | |
1962 | } | |
1963 | ||
1964 | int kvm_device_ioctl(int fd, int type, ...) | |
1965 | { | |
1966 | int ret; | |
1967 | void *arg; | |
1968 | va_list ap; | |
1969 | ||
1970 | va_start(ap, type); | |
1971 | arg = va_arg(ap, void *); | |
1972 | va_end(ap); | |
1973 | ||
1974 | trace_kvm_device_ioctl(fd, type, arg); | |
1975 | ret = ioctl(fd, type, arg); | |
1976 | if (ret == -1) { | |
1977 | ret = -errno; | |
1978 | } | |
1979 | return ret; | |
1980 | } | |
1981 | ||
1982 | int kvm_vm_check_attr(KVMState *s, uint32_t group, uint64_t attr) | |
1983 | { | |
1984 | int ret; | |
1985 | struct kvm_device_attr attribute = { | |
1986 | .group = group, | |
1987 | .attr = attr, | |
1988 | }; | |
1989 | ||
1990 | if (!kvm_vm_attributes_allowed) { | |
1991 | return 0; | |
1992 | } | |
1993 | ||
1994 | ret = kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attribute); | |
1995 | /* kvm returns 0 on success for HAS_DEVICE_ATTR */ | |
1996 | return ret ? 0 : 1; | |
1997 | } | |
1998 | ||
1999 | int kvm_device_check_attr(int dev_fd, uint32_t group, uint64_t attr) | |
2000 | { | |
2001 | struct kvm_device_attr attribute = { | |
2002 | .group = group, | |
2003 | .attr = attr, | |
2004 | .flags = 0, | |
2005 | }; | |
2006 | ||
2007 | return kvm_device_ioctl(dev_fd, KVM_HAS_DEVICE_ATTR, &attribute) ? 0 : 1; | |
2008 | } | |
2009 | ||
2010 | void kvm_device_access(int fd, int group, uint64_t attr, | |
2011 | void *val, bool write) | |
2012 | { | |
2013 | struct kvm_device_attr kvmattr; | |
2014 | int err; | |
2015 | ||
2016 | kvmattr.flags = 0; | |
2017 | kvmattr.group = group; | |
2018 | kvmattr.attr = attr; | |
2019 | kvmattr.addr = (uintptr_t)val; | |
2020 | ||
2021 | err = kvm_device_ioctl(fd, | |
2022 | write ? KVM_SET_DEVICE_ATTR : KVM_GET_DEVICE_ATTR, | |
2023 | &kvmattr); | |
2024 | if (err < 0) { | |
2025 | error_report("KVM_%s_DEVICE_ATTR failed: %s\n" | |
2026 | "Group %d attr 0x%016" PRIx64, write ? "SET" : "GET", | |
2027 | strerror(-err), group, attr); | |
2028 | abort(); | |
2029 | } | |
2030 | } | |
2031 | ||
2032 | int kvm_has_sync_mmu(void) | |
2033 | { | |
2034 | return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU); | |
2035 | } | |
2036 | ||
2037 | int kvm_has_vcpu_events(void) | |
2038 | { | |
2039 | return kvm_state->vcpu_events; | |
2040 | } | |
2041 | ||
2042 | int kvm_has_robust_singlestep(void) | |
2043 | { | |
2044 | return kvm_state->robust_singlestep; | |
2045 | } | |
2046 | ||
2047 | int kvm_has_debugregs(void) | |
2048 | { | |
2049 | return kvm_state->debugregs; | |
2050 | } | |
2051 | ||
2052 | int kvm_has_many_ioeventfds(void) | |
2053 | { | |
2054 | if (!kvm_enabled()) { | |
2055 | return 0; | |
2056 | } | |
2057 | return kvm_state->many_ioeventfds; | |
2058 | } | |
2059 | ||
2060 | int kvm_has_gsi_routing(void) | |
2061 | { | |
2062 | #ifdef KVM_CAP_IRQ_ROUTING | |
2063 | return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING); | |
2064 | #else | |
2065 | return false; | |
2066 | #endif | |
2067 | } | |
2068 | ||
2069 | int kvm_has_intx_set_mask(void) | |
2070 | { | |
2071 | return kvm_state->intx_set_mask; | |
2072 | } | |
2073 | ||
2074 | void kvm_setup_guest_memory(void *start, size_t size) | |
2075 | { | |
2076 | if (!kvm_has_sync_mmu()) { | |
2077 | int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK); | |
2078 | ||
2079 | if (ret) { | |
2080 | perror("qemu_madvise"); | |
2081 | fprintf(stderr, | |
2082 | "Need MADV_DONTFORK in absence of synchronous KVM MMU\n"); | |
2083 | exit(1); | |
2084 | } | |
2085 | } | |
2086 | } | |
2087 | ||
2088 | #ifdef KVM_CAP_SET_GUEST_DEBUG | |
2089 | struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu, | |
2090 | target_ulong pc) | |
2091 | { | |
2092 | struct kvm_sw_breakpoint *bp; | |
2093 | ||
2094 | QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) { | |
2095 | if (bp->pc == pc) { | |
2096 | return bp; | |
2097 | } | |
2098 | } | |
2099 | return NULL; | |
2100 | } | |
2101 | ||
2102 | int kvm_sw_breakpoints_active(CPUState *cpu) | |
2103 | { | |
2104 | return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints); | |
2105 | } | |
2106 | ||
2107 | struct kvm_set_guest_debug_data { | |
2108 | struct kvm_guest_debug dbg; | |
2109 | CPUState *cpu; | |
2110 | int err; | |
2111 | }; | |
2112 | ||
2113 | static void kvm_invoke_set_guest_debug(void *data) | |
2114 | { | |
2115 | struct kvm_set_guest_debug_data *dbg_data = data; | |
2116 | ||
2117 | dbg_data->err = kvm_vcpu_ioctl(dbg_data->cpu, KVM_SET_GUEST_DEBUG, | |
2118 | &dbg_data->dbg); | |
2119 | } | |
2120 | ||
2121 | int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap) | |
2122 | { | |
2123 | struct kvm_set_guest_debug_data data; | |
2124 | ||
2125 | data.dbg.control = reinject_trap; | |
2126 | ||
2127 | if (cpu->singlestep_enabled) { | |
2128 | data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP; | |
2129 | } | |
2130 | kvm_arch_update_guest_debug(cpu, &data.dbg); | |
2131 | data.cpu = cpu; | |
2132 | ||
2133 | run_on_cpu(cpu, kvm_invoke_set_guest_debug, &data); | |
2134 | return data.err; | |
2135 | } | |
2136 | ||
2137 | int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr, | |
2138 | target_ulong len, int type) | |
2139 | { | |
2140 | struct kvm_sw_breakpoint *bp; | |
2141 | int err; | |
2142 | ||
2143 | if (type == GDB_BREAKPOINT_SW) { | |
2144 | bp = kvm_find_sw_breakpoint(cpu, addr); | |
2145 | if (bp) { | |
2146 | bp->use_count++; | |
2147 | return 0; | |
2148 | } | |
2149 | ||
2150 | bp = g_malloc(sizeof(struct kvm_sw_breakpoint)); | |
2151 | bp->pc = addr; | |
2152 | bp->use_count = 1; | |
2153 | err = kvm_arch_insert_sw_breakpoint(cpu, bp); | |
2154 | if (err) { | |
2155 | g_free(bp); | |
2156 | return err; | |
2157 | } | |
2158 | ||
2159 | QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry); | |
2160 | } else { | |
2161 | err = kvm_arch_insert_hw_breakpoint(addr, len, type); | |
2162 | if (err) { | |
2163 | return err; | |
2164 | } | |
2165 | } | |
2166 | ||
2167 | CPU_FOREACH(cpu) { | |
2168 | err = kvm_update_guest_debug(cpu, 0); | |
2169 | if (err) { | |
2170 | return err; | |
2171 | } | |
2172 | } | |
2173 | return 0; | |
2174 | } | |
2175 | ||
2176 | int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr, | |
2177 | target_ulong len, int type) | |
2178 | { | |
2179 | struct kvm_sw_breakpoint *bp; | |
2180 | int err; | |
2181 | ||
2182 | if (type == GDB_BREAKPOINT_SW) { | |
2183 | bp = kvm_find_sw_breakpoint(cpu, addr); | |
2184 | if (!bp) { | |
2185 | return -ENOENT; | |
2186 | } | |
2187 | ||
2188 | if (bp->use_count > 1) { | |
2189 | bp->use_count--; | |
2190 | return 0; | |
2191 | } | |
2192 | ||
2193 | err = kvm_arch_remove_sw_breakpoint(cpu, bp); | |
2194 | if (err) { | |
2195 | return err; | |
2196 | } | |
2197 | ||
2198 | QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry); | |
2199 | g_free(bp); | |
2200 | } else { | |
2201 | err = kvm_arch_remove_hw_breakpoint(addr, len, type); | |
2202 | if (err) { | |
2203 | return err; | |
2204 | } | |
2205 | } | |
2206 | ||
2207 | CPU_FOREACH(cpu) { | |
2208 | err = kvm_update_guest_debug(cpu, 0); | |
2209 | if (err) { | |
2210 | return err; | |
2211 | } | |
2212 | } | |
2213 | return 0; | |
2214 | } | |
2215 | ||
2216 | void kvm_remove_all_breakpoints(CPUState *cpu) | |
2217 | { | |
2218 | struct kvm_sw_breakpoint *bp, *next; | |
2219 | KVMState *s = cpu->kvm_state; | |
2220 | CPUState *tmpcpu; | |
2221 | ||
2222 | QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) { | |
2223 | if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) { | |
2224 | /* Try harder to find a CPU that currently sees the breakpoint. */ | |
2225 | CPU_FOREACH(tmpcpu) { | |
2226 | if (kvm_arch_remove_sw_breakpoint(tmpcpu, bp) == 0) { | |
2227 | break; | |
2228 | } | |
2229 | } | |
2230 | } | |
2231 | QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry); | |
2232 | g_free(bp); | |
2233 | } | |
2234 | kvm_arch_remove_all_hw_breakpoints(); | |
2235 | ||
2236 | CPU_FOREACH(cpu) { | |
2237 | kvm_update_guest_debug(cpu, 0); | |
2238 | } | |
2239 | } | |
2240 | ||
2241 | #else /* !KVM_CAP_SET_GUEST_DEBUG */ | |
2242 | ||
2243 | int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap) | |
2244 | { | |
2245 | return -EINVAL; | |
2246 | } | |
2247 | ||
2248 | int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr, | |
2249 | target_ulong len, int type) | |
2250 | { | |
2251 | return -EINVAL; | |
2252 | } | |
2253 | ||
2254 | int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr, | |
2255 | target_ulong len, int type) | |
2256 | { | |
2257 | return -EINVAL; | |
2258 | } | |
2259 | ||
2260 | void kvm_remove_all_breakpoints(CPUState *cpu) | |
2261 | { | |
2262 | } | |
2263 | #endif /* !KVM_CAP_SET_GUEST_DEBUG */ | |
2264 | ||
2265 | int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset) | |
2266 | { | |
2267 | KVMState *s = kvm_state; | |
2268 | struct kvm_signal_mask *sigmask; | |
2269 | int r; | |
2270 | ||
2271 | if (!sigset) { | |
2272 | return kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, NULL); | |
2273 | } | |
2274 | ||
2275 | sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset)); | |
2276 | ||
2277 | sigmask->len = s->sigmask_len; | |
2278 | memcpy(sigmask->sigset, sigset, sizeof(*sigset)); | |
2279 | r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask); | |
2280 | g_free(sigmask); | |
2281 | ||
2282 | return r; | |
2283 | } | |
2284 | int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr) | |
2285 | { | |
2286 | return kvm_arch_on_sigbus_vcpu(cpu, code, addr); | |
2287 | } | |
2288 | ||
2289 | int kvm_on_sigbus(int code, void *addr) | |
2290 | { | |
2291 | return kvm_arch_on_sigbus(code, addr); | |
2292 | } | |
2293 | ||
2294 | int kvm_create_device(KVMState *s, uint64_t type, bool test) | |
2295 | { | |
2296 | int ret; | |
2297 | struct kvm_create_device create_dev; | |
2298 | ||
2299 | create_dev.type = type; | |
2300 | create_dev.fd = -1; | |
2301 | create_dev.flags = test ? KVM_CREATE_DEVICE_TEST : 0; | |
2302 | ||
2303 | if (!kvm_check_extension(s, KVM_CAP_DEVICE_CTRL)) { | |
2304 | return -ENOTSUP; | |
2305 | } | |
2306 | ||
2307 | ret = kvm_vm_ioctl(s, KVM_CREATE_DEVICE, &create_dev); | |
2308 | if (ret) { | |
2309 | return ret; | |
2310 | } | |
2311 | ||
2312 | return test ? 0 : create_dev.fd; | |
2313 | } | |
2314 | ||
2315 | int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source) | |
2316 | { | |
2317 | struct kvm_one_reg reg; | |
2318 | int r; | |
2319 | ||
2320 | reg.id = id; | |
2321 | reg.addr = (uintptr_t) source; | |
2322 | r = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®); | |
2323 | if (r) { | |
2324 | trace_kvm_failed_reg_set(id, strerror(r)); | |
2325 | } | |
2326 | return r; | |
2327 | } | |
2328 | ||
2329 | int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target) | |
2330 | { | |
2331 | struct kvm_one_reg reg; | |
2332 | int r; | |
2333 | ||
2334 | reg.id = id; | |
2335 | reg.addr = (uintptr_t) target; | |
2336 | r = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, ®); | |
2337 | if (r) { | |
2338 | trace_kvm_failed_reg_get(id, strerror(r)); | |
2339 | } | |
2340 | return r; | |
2341 | } | |
2342 | ||
2343 | static void kvm_accel_class_init(ObjectClass *oc, void *data) | |
2344 | { | |
2345 | AccelClass *ac = ACCEL_CLASS(oc); | |
2346 | ac->name = "KVM"; | |
2347 | ac->init_machine = kvm_init; | |
2348 | ac->allowed = &kvm_allowed; | |
2349 | } | |
2350 | ||
2351 | static const TypeInfo kvm_accel_type = { | |
2352 | .name = TYPE_KVM_ACCEL, | |
2353 | .parent = TYPE_ACCEL, | |
2354 | .class_init = kvm_accel_class_init, | |
2355 | .instance_size = sizeof(KVMState), | |
2356 | }; | |
2357 | ||
2358 | static void kvm_type_init(void) | |
2359 | { | |
2360 | type_register_static(&kvm_accel_type); | |
2361 | } | |
2362 | ||
2363 | type_init(kvm_type_init); |