fdc: use IsaDma interface instead of global DMA_* functions
[qemu/ar7.git] / kvm-all.c
blob914888992119783d47fa69db80a5251cf623028e
1 /*
2 * QEMU KVM support
4 * Copyright IBM, Corp. 2008
5 * Red Hat, Inc. 2008
7 * Authors:
8 * Anthony Liguori <aliguori@us.ibm.com>
9 * Glauber Costa <gcosta@redhat.com>
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.
16 #include <sys/types.h>
17 #include <sys/ioctl.h>
18 #include <sys/mman.h>
19 #include <stdarg.h>
21 #include <linux/kvm.h>
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"
41 #include "hw/boards.h"
43 /* This check must be after config-host.h is included */
44 #ifdef CONFIG_EVENTFD
45 #include <sys/eventfd.h>
46 #endif
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.
51 #define PAGE_SIZE getpagesize()
53 //#define DEBUG_KVM
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
63 #define KVM_MSI_HASHTAB_SIZE 256
65 struct KVMState
67 AccelState parent_obj;
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;
100 KVMState *kvm_state;
101 bool kvm_kernel_irqchip;
102 bool kvm_split_irqchip;
103 bool kvm_async_interrupts_allowed;
104 bool kvm_halt_in_kernel_allowed;
105 bool kvm_eventfds_allowed;
106 bool kvm_irqfds_allowed;
107 bool kvm_resamplefds_allowed;
108 bool kvm_msi_via_irqfd_allowed;
109 bool kvm_gsi_routing_allowed;
110 bool kvm_gsi_direct_mapping;
111 bool kvm_allowed;
112 bool kvm_readonly_mem_allowed;
113 bool kvm_vm_attributes_allowed;
114 bool kvm_direct_msi_allowed;
115 bool kvm_ioeventfd_any_length_allowed;
117 static const KVMCapabilityInfo kvm_required_capabilites[] = {
118 KVM_CAP_INFO(USER_MEMORY),
119 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS),
120 KVM_CAP_LAST_INFO
123 static KVMSlot *kvm_get_free_slot(KVMMemoryListener *kml)
125 KVMState *s = kvm_state;
126 int i;
128 for (i = 0; i < s->nr_slots; i++) {
129 if (kml->slots[i].memory_size == 0) {
130 return &kml->slots[i];
134 return NULL;
137 bool kvm_has_free_slot(MachineState *ms)
139 KVMState *s = KVM_STATE(ms->accelerator);
141 return kvm_get_free_slot(&s->memory_listener);
144 static KVMSlot *kvm_alloc_slot(KVMMemoryListener *kml)
146 KVMSlot *slot = kvm_get_free_slot(kml);
148 if (slot) {
149 return slot;
152 fprintf(stderr, "%s: no free slot available\n", __func__);
153 abort();
156 static KVMSlot *kvm_lookup_matching_slot(KVMMemoryListener *kml,
157 hwaddr start_addr,
158 hwaddr end_addr)
160 KVMState *s = kvm_state;
161 int i;
163 for (i = 0; i < s->nr_slots; i++) {
164 KVMSlot *mem = &kml->slots[i];
166 if (start_addr == mem->start_addr &&
167 end_addr == mem->start_addr + mem->memory_size) {
168 return mem;
172 return NULL;
176 * Find overlapping slot with lowest start address
178 static KVMSlot *kvm_lookup_overlapping_slot(KVMMemoryListener *kml,
179 hwaddr start_addr,
180 hwaddr end_addr)
182 KVMState *s = kvm_state;
183 KVMSlot *found = NULL;
184 int i;
186 for (i = 0; i < s->nr_slots; i++) {
187 KVMSlot *mem = &kml->slots[i];
189 if (mem->memory_size == 0 ||
190 (found && found->start_addr < mem->start_addr)) {
191 continue;
194 if (end_addr > mem->start_addr &&
195 start_addr < mem->start_addr + mem->memory_size) {
196 found = mem;
200 return found;
203 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram,
204 hwaddr *phys_addr)
206 KVMMemoryListener *kml = &s->memory_listener;
207 int i;
209 for (i = 0; i < s->nr_slots; i++) {
210 KVMSlot *mem = &kml->slots[i];
212 if (ram >= mem->ram && ram < mem->ram + mem->memory_size) {
213 *phys_addr = mem->start_addr + (ram - mem->ram);
214 return 1;
218 return 0;
221 static int kvm_set_user_memory_region(KVMMemoryListener *kml, KVMSlot *slot)
223 KVMState *s = kvm_state;
224 struct kvm_userspace_memory_region mem;
226 mem.slot = slot->slot | (kml->as_id << 16);
227 mem.guest_phys_addr = slot->start_addr;
228 mem.userspace_addr = (unsigned long)slot->ram;
229 mem.flags = slot->flags;
231 if (slot->memory_size && mem.flags & KVM_MEM_READONLY) {
232 /* Set the slot size to 0 before setting the slot to the desired
233 * value. This is needed based on KVM commit 75d61fbc. */
234 mem.memory_size = 0;
235 kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
237 mem.memory_size = slot->memory_size;
238 return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
241 int kvm_init_vcpu(CPUState *cpu)
243 KVMState *s = kvm_state;
244 long mmap_size;
245 int ret;
247 DPRINTF("kvm_init_vcpu\n");
249 ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)kvm_arch_vcpu_id(cpu));
250 if (ret < 0) {
251 DPRINTF("kvm_create_vcpu failed\n");
252 goto err;
255 cpu->kvm_fd = ret;
256 cpu->kvm_state = s;
257 cpu->kvm_vcpu_dirty = true;
259 mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
260 if (mmap_size < 0) {
261 ret = mmap_size;
262 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
263 goto err;
266 cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
267 cpu->kvm_fd, 0);
268 if (cpu->kvm_run == MAP_FAILED) {
269 ret = -errno;
270 DPRINTF("mmap'ing vcpu state failed\n");
271 goto err;
274 if (s->coalesced_mmio && !s->coalesced_mmio_ring) {
275 s->coalesced_mmio_ring =
276 (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE;
279 ret = kvm_arch_init_vcpu(cpu);
280 err:
281 return ret;
285 * dirty pages logging control
288 static int kvm_mem_flags(MemoryRegion *mr)
290 bool readonly = mr->readonly || memory_region_is_romd(mr);
291 int flags = 0;
293 if (memory_region_get_dirty_log_mask(mr) != 0) {
294 flags |= KVM_MEM_LOG_DIRTY_PAGES;
296 if (readonly && kvm_readonly_mem_allowed) {
297 flags |= KVM_MEM_READONLY;
299 return flags;
302 static int kvm_slot_update_flags(KVMMemoryListener *kml, KVMSlot *mem,
303 MemoryRegion *mr)
305 int old_flags;
307 old_flags = mem->flags;
308 mem->flags = kvm_mem_flags(mr);
310 /* If nothing changed effectively, no need to issue ioctl */
311 if (mem->flags == old_flags) {
312 return 0;
315 return kvm_set_user_memory_region(kml, mem);
318 static int kvm_section_update_flags(KVMMemoryListener *kml,
319 MemoryRegionSection *section)
321 hwaddr phys_addr = section->offset_within_address_space;
322 ram_addr_t size = int128_get64(section->size);
323 KVMSlot *mem = kvm_lookup_matching_slot(kml, phys_addr, phys_addr + size);
325 if (mem == NULL) {
326 return 0;
327 } else {
328 return kvm_slot_update_flags(kml, mem, section->mr);
332 static void kvm_log_start(MemoryListener *listener,
333 MemoryRegionSection *section,
334 int old, int new)
336 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
337 int r;
339 if (old != 0) {
340 return;
343 r = kvm_section_update_flags(kml, section);
344 if (r < 0) {
345 abort();
349 static void kvm_log_stop(MemoryListener *listener,
350 MemoryRegionSection *section,
351 int old, int new)
353 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
354 int r;
356 if (new != 0) {
357 return;
360 r = kvm_section_update_flags(kml, section);
361 if (r < 0) {
362 abort();
366 /* get kvm's dirty pages bitmap and update qemu's */
367 static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section,
368 unsigned long *bitmap)
370 ram_addr_t start = section->offset_within_region + section->mr->ram_addr;
371 ram_addr_t pages = int128_get64(section->size) / getpagesize();
373 cpu_physical_memory_set_dirty_lebitmap(bitmap, start, pages);
374 return 0;
377 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
380 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
381 * This function updates qemu's dirty bitmap using
382 * memory_region_set_dirty(). This means all bits are set
383 * to dirty.
385 * @start_add: start of logged region.
386 * @end_addr: end of logged region.
388 static int kvm_physical_sync_dirty_bitmap(KVMMemoryListener *kml,
389 MemoryRegionSection *section)
391 KVMState *s = kvm_state;
392 unsigned long size, allocated_size = 0;
393 struct kvm_dirty_log d = {};
394 KVMSlot *mem;
395 int ret = 0;
396 hwaddr start_addr = section->offset_within_address_space;
397 hwaddr end_addr = start_addr + int128_get64(section->size);
399 d.dirty_bitmap = NULL;
400 while (start_addr < end_addr) {
401 mem = kvm_lookup_overlapping_slot(kml, start_addr, end_addr);
402 if (mem == NULL) {
403 break;
406 /* XXX bad kernel interface alert
407 * For dirty bitmap, kernel allocates array of size aligned to
408 * bits-per-long. But for case when the kernel is 64bits and
409 * the userspace is 32bits, userspace can't align to the same
410 * bits-per-long, since sizeof(long) is different between kernel
411 * and user space. This way, userspace will provide buffer which
412 * may be 4 bytes less than the kernel will use, resulting in
413 * userspace memory corruption (which is not detectable by valgrind
414 * too, in most cases).
415 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
416 * a hope that sizeof(long) wont become >8 any time soon.
418 size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS),
419 /*HOST_LONG_BITS*/ 64) / 8;
420 if (!d.dirty_bitmap) {
421 d.dirty_bitmap = g_malloc(size);
422 } else if (size > allocated_size) {
423 d.dirty_bitmap = g_realloc(d.dirty_bitmap, size);
425 allocated_size = size;
426 memset(d.dirty_bitmap, 0, allocated_size);
428 d.slot = mem->slot | (kml->as_id << 16);
429 if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
430 DPRINTF("ioctl failed %d\n", errno);
431 ret = -1;
432 break;
435 kvm_get_dirty_pages_log_range(section, d.dirty_bitmap);
436 start_addr = mem->start_addr + mem->memory_size;
438 g_free(d.dirty_bitmap);
440 return ret;
443 static void kvm_coalesce_mmio_region(MemoryListener *listener,
444 MemoryRegionSection *secion,
445 hwaddr start, hwaddr size)
447 KVMState *s = kvm_state;
449 if (s->coalesced_mmio) {
450 struct kvm_coalesced_mmio_zone zone;
452 zone.addr = start;
453 zone.size = size;
454 zone.pad = 0;
456 (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
460 static void kvm_uncoalesce_mmio_region(MemoryListener *listener,
461 MemoryRegionSection *secion,
462 hwaddr start, hwaddr size)
464 KVMState *s = kvm_state;
466 if (s->coalesced_mmio) {
467 struct kvm_coalesced_mmio_zone zone;
469 zone.addr = start;
470 zone.size = size;
471 zone.pad = 0;
473 (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
477 int kvm_check_extension(KVMState *s, unsigned int extension)
479 int ret;
481 ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
482 if (ret < 0) {
483 ret = 0;
486 return ret;
489 int kvm_vm_check_extension(KVMState *s, unsigned int extension)
491 int ret;
493 ret = kvm_vm_ioctl(s, KVM_CHECK_EXTENSION, extension);
494 if (ret < 0) {
495 /* VM wide version not implemented, use global one instead */
496 ret = kvm_check_extension(s, extension);
499 return ret;
502 static uint32_t adjust_ioeventfd_endianness(uint32_t val, uint32_t size)
504 #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN)
505 /* The kernel expects ioeventfd values in HOST_WORDS_BIGENDIAN
506 * endianness, but the memory core hands them in target endianness.
507 * For example, PPC is always treated as big-endian even if running
508 * on KVM and on PPC64LE. Correct here.
510 switch (size) {
511 case 2:
512 val = bswap16(val);
513 break;
514 case 4:
515 val = bswap32(val);
516 break;
518 #endif
519 return val;
522 static int kvm_set_ioeventfd_mmio(int fd, hwaddr addr, uint32_t val,
523 bool assign, uint32_t size, bool datamatch)
525 int ret;
526 struct kvm_ioeventfd iofd = {
527 .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0,
528 .addr = addr,
529 .len = size,
530 .flags = 0,
531 .fd = fd,
534 if (!kvm_enabled()) {
535 return -ENOSYS;
538 if (datamatch) {
539 iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
541 if (!assign) {
542 iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
545 ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd);
547 if (ret < 0) {
548 return -errno;
551 return 0;
554 static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val,
555 bool assign, uint32_t size, bool datamatch)
557 struct kvm_ioeventfd kick = {
558 .datamatch = datamatch ? adjust_ioeventfd_endianness(val, size) : 0,
559 .addr = addr,
560 .flags = KVM_IOEVENTFD_FLAG_PIO,
561 .len = size,
562 .fd = fd,
564 int r;
565 if (!kvm_enabled()) {
566 return -ENOSYS;
568 if (datamatch) {
569 kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
571 if (!assign) {
572 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
574 r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
575 if (r < 0) {
576 return r;
578 return 0;
582 static int kvm_check_many_ioeventfds(void)
584 /* Userspace can use ioeventfd for io notification. This requires a host
585 * that supports eventfd(2) and an I/O thread; since eventfd does not
586 * support SIGIO it cannot interrupt the vcpu.
588 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
589 * can avoid creating too many ioeventfds.
591 #if defined(CONFIG_EVENTFD)
592 int ioeventfds[7];
593 int i, ret = 0;
594 for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) {
595 ioeventfds[i] = eventfd(0, EFD_CLOEXEC);
596 if (ioeventfds[i] < 0) {
597 break;
599 ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true);
600 if (ret < 0) {
601 close(ioeventfds[i]);
602 break;
606 /* Decide whether many devices are supported or not */
607 ret = i == ARRAY_SIZE(ioeventfds);
609 while (i-- > 0) {
610 kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true);
611 close(ioeventfds[i]);
613 return ret;
614 #else
615 return 0;
616 #endif
619 static const KVMCapabilityInfo *
620 kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list)
622 while (list->name) {
623 if (!kvm_check_extension(s, list->value)) {
624 return list;
626 list++;
628 return NULL;
631 static void kvm_set_phys_mem(KVMMemoryListener *kml,
632 MemoryRegionSection *section, bool add)
634 KVMState *s = kvm_state;
635 KVMSlot *mem, old;
636 int err;
637 MemoryRegion *mr = section->mr;
638 bool writeable = !mr->readonly && !mr->rom_device;
639 hwaddr start_addr = section->offset_within_address_space;
640 ram_addr_t size = int128_get64(section->size);
641 void *ram = NULL;
642 unsigned delta;
644 /* kvm works in page size chunks, but the function may be called
645 with sub-page size and unaligned start address. Pad the start
646 address to next and truncate size to previous page boundary. */
647 delta = qemu_real_host_page_size - (start_addr & ~qemu_real_host_page_mask);
648 delta &= ~qemu_real_host_page_mask;
649 if (delta > size) {
650 return;
652 start_addr += delta;
653 size -= delta;
654 size &= qemu_real_host_page_mask;
655 if (!size || (start_addr & ~qemu_real_host_page_mask)) {
656 return;
659 if (!memory_region_is_ram(mr)) {
660 if (writeable || !kvm_readonly_mem_allowed) {
661 return;
662 } else if (!mr->romd_mode) {
663 /* If the memory device is not in romd_mode, then we actually want
664 * to remove the kvm memory slot so all accesses will trap. */
665 add = false;
669 ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta;
671 while (1) {
672 mem = kvm_lookup_overlapping_slot(kml, start_addr, start_addr + size);
673 if (!mem) {
674 break;
677 if (add && start_addr >= mem->start_addr &&
678 (start_addr + size <= mem->start_addr + mem->memory_size) &&
679 (ram - start_addr == mem->ram - mem->start_addr)) {
680 /* The new slot fits into the existing one and comes with
681 * identical parameters - update flags and done. */
682 kvm_slot_update_flags(kml, mem, mr);
683 return;
686 old = *mem;
688 if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
689 kvm_physical_sync_dirty_bitmap(kml, section);
692 /* unregister the overlapping slot */
693 mem->memory_size = 0;
694 err = kvm_set_user_memory_region(kml, mem);
695 if (err) {
696 fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
697 __func__, strerror(-err));
698 abort();
701 /* Workaround for older KVM versions: we can't join slots, even not by
702 * unregistering the previous ones and then registering the larger
703 * slot. We have to maintain the existing fragmentation. Sigh.
705 * This workaround assumes that the new slot starts at the same
706 * address as the first existing one. If not or if some overlapping
707 * slot comes around later, we will fail (not seen in practice so far)
708 * - and actually require a recent KVM version. */
709 if (s->broken_set_mem_region &&
710 old.start_addr == start_addr && old.memory_size < size && add) {
711 mem = kvm_alloc_slot(kml);
712 mem->memory_size = old.memory_size;
713 mem->start_addr = old.start_addr;
714 mem->ram = old.ram;
715 mem->flags = kvm_mem_flags(mr);
717 err = kvm_set_user_memory_region(kml, mem);
718 if (err) {
719 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
720 strerror(-err));
721 abort();
724 start_addr += old.memory_size;
725 ram += old.memory_size;
726 size -= old.memory_size;
727 continue;
730 /* register prefix slot */
731 if (old.start_addr < start_addr) {
732 mem = kvm_alloc_slot(kml);
733 mem->memory_size = start_addr - old.start_addr;
734 mem->start_addr = old.start_addr;
735 mem->ram = old.ram;
736 mem->flags = kvm_mem_flags(mr);
738 err = kvm_set_user_memory_region(kml, mem);
739 if (err) {
740 fprintf(stderr, "%s: error registering prefix slot: %s\n",
741 __func__, strerror(-err));
742 #ifdef TARGET_PPC
743 fprintf(stderr, "%s: This is probably because your kernel's " \
744 "PAGE_SIZE is too big. Please try to use 4k " \
745 "PAGE_SIZE!\n", __func__);
746 #endif
747 abort();
751 /* register suffix slot */
752 if (old.start_addr + old.memory_size > start_addr + size) {
753 ram_addr_t size_delta;
755 mem = kvm_alloc_slot(kml);
756 mem->start_addr = start_addr + size;
757 size_delta = mem->start_addr - old.start_addr;
758 mem->memory_size = old.memory_size - size_delta;
759 mem->ram = old.ram + size_delta;
760 mem->flags = kvm_mem_flags(mr);
762 err = kvm_set_user_memory_region(kml, mem);
763 if (err) {
764 fprintf(stderr, "%s: error registering suffix slot: %s\n",
765 __func__, strerror(-err));
766 abort();
771 /* in case the KVM bug workaround already "consumed" the new slot */
772 if (!size) {
773 return;
775 if (!add) {
776 return;
778 mem = kvm_alloc_slot(kml);
779 mem->memory_size = size;
780 mem->start_addr = start_addr;
781 mem->ram = ram;
782 mem->flags = kvm_mem_flags(mr);
784 err = kvm_set_user_memory_region(kml, mem);
785 if (err) {
786 fprintf(stderr, "%s: error registering slot: %s\n", __func__,
787 strerror(-err));
788 abort();
792 static void kvm_region_add(MemoryListener *listener,
793 MemoryRegionSection *section)
795 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
797 memory_region_ref(section->mr);
798 kvm_set_phys_mem(kml, section, true);
801 static void kvm_region_del(MemoryListener *listener,
802 MemoryRegionSection *section)
804 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
806 kvm_set_phys_mem(kml, section, false);
807 memory_region_unref(section->mr);
810 static void kvm_log_sync(MemoryListener *listener,
811 MemoryRegionSection *section)
813 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
814 int r;
816 r = kvm_physical_sync_dirty_bitmap(kml, section);
817 if (r < 0) {
818 abort();
822 static void kvm_mem_ioeventfd_add(MemoryListener *listener,
823 MemoryRegionSection *section,
824 bool match_data, uint64_t data,
825 EventNotifier *e)
827 int fd = event_notifier_get_fd(e);
828 int r;
830 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
831 data, true, int128_get64(section->size),
832 match_data);
833 if (r < 0) {
834 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
835 __func__, strerror(-r));
836 abort();
840 static void kvm_mem_ioeventfd_del(MemoryListener *listener,
841 MemoryRegionSection *section,
842 bool match_data, uint64_t data,
843 EventNotifier *e)
845 int fd = event_notifier_get_fd(e);
846 int r;
848 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
849 data, false, int128_get64(section->size),
850 match_data);
851 if (r < 0) {
852 abort();
856 static void kvm_io_ioeventfd_add(MemoryListener *listener,
857 MemoryRegionSection *section,
858 bool match_data, uint64_t data,
859 EventNotifier *e)
861 int fd = event_notifier_get_fd(e);
862 int r;
864 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
865 data, true, int128_get64(section->size),
866 match_data);
867 if (r < 0) {
868 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
869 __func__, strerror(-r));
870 abort();
874 static void kvm_io_ioeventfd_del(MemoryListener *listener,
875 MemoryRegionSection *section,
876 bool match_data, uint64_t data,
877 EventNotifier *e)
880 int fd = event_notifier_get_fd(e);
881 int r;
883 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
884 data, false, int128_get64(section->size),
885 match_data);
886 if (r < 0) {
887 abort();
891 void kvm_memory_listener_register(KVMState *s, KVMMemoryListener *kml,
892 AddressSpace *as, int as_id)
894 int i;
896 kml->slots = g_malloc0(s->nr_slots * sizeof(KVMSlot));
897 kml->as_id = as_id;
899 for (i = 0; i < s->nr_slots; i++) {
900 kml->slots[i].slot = i;
903 kml->listener.region_add = kvm_region_add;
904 kml->listener.region_del = kvm_region_del;
905 kml->listener.log_start = kvm_log_start;
906 kml->listener.log_stop = kvm_log_stop;
907 kml->listener.log_sync = kvm_log_sync;
908 kml->listener.priority = 10;
910 memory_listener_register(&kml->listener, as);
913 static MemoryListener kvm_io_listener = {
914 .eventfd_add = kvm_io_ioeventfd_add,
915 .eventfd_del = kvm_io_ioeventfd_del,
916 .priority = 10,
919 static void kvm_handle_interrupt(CPUState *cpu, int mask)
921 cpu->interrupt_request |= mask;
923 if (!qemu_cpu_is_self(cpu)) {
924 qemu_cpu_kick(cpu);
928 int kvm_set_irq(KVMState *s, int irq, int level)
930 struct kvm_irq_level event;
931 int ret;
933 assert(kvm_async_interrupts_enabled());
935 event.level = level;
936 event.irq = irq;
937 ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event);
938 if (ret < 0) {
939 perror("kvm_set_irq");
940 abort();
943 return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status;
946 #ifdef KVM_CAP_IRQ_ROUTING
947 typedef struct KVMMSIRoute {
948 struct kvm_irq_routing_entry kroute;
949 QTAILQ_ENTRY(KVMMSIRoute) entry;
950 } KVMMSIRoute;
952 static void set_gsi(KVMState *s, unsigned int gsi)
954 s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32);
957 static void clear_gsi(KVMState *s, unsigned int gsi)
959 s->used_gsi_bitmap[gsi / 32] &= ~(1U << (gsi % 32));
962 void kvm_init_irq_routing(KVMState *s)
964 int gsi_count, i;
966 gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING) - 1;
967 if (gsi_count > 0) {
968 unsigned int gsi_bits, i;
970 /* Round up so we can search ints using ffs */
971 gsi_bits = ALIGN(gsi_count, 32);
972 s->used_gsi_bitmap = g_malloc0(gsi_bits / 8);
973 s->gsi_count = gsi_count;
975 /* Mark any over-allocated bits as already in use */
976 for (i = gsi_count; i < gsi_bits; i++) {
977 set_gsi(s, i);
981 s->irq_routes = g_malloc0(sizeof(*s->irq_routes));
982 s->nr_allocated_irq_routes = 0;
984 if (!kvm_direct_msi_allowed) {
985 for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) {
986 QTAILQ_INIT(&s->msi_hashtab[i]);
990 kvm_arch_init_irq_routing(s);
993 void kvm_irqchip_commit_routes(KVMState *s)
995 int ret;
997 s->irq_routes->flags = 0;
998 ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes);
999 assert(ret == 0);
1002 static void kvm_add_routing_entry(KVMState *s,
1003 struct kvm_irq_routing_entry *entry)
1005 struct kvm_irq_routing_entry *new;
1006 int n, size;
1008 if (s->irq_routes->nr == s->nr_allocated_irq_routes) {
1009 n = s->nr_allocated_irq_routes * 2;
1010 if (n < 64) {
1011 n = 64;
1013 size = sizeof(struct kvm_irq_routing);
1014 size += n * sizeof(*new);
1015 s->irq_routes = g_realloc(s->irq_routes, size);
1016 s->nr_allocated_irq_routes = n;
1018 n = s->irq_routes->nr++;
1019 new = &s->irq_routes->entries[n];
1021 *new = *entry;
1023 set_gsi(s, entry->gsi);
1026 static int kvm_update_routing_entry(KVMState *s,
1027 struct kvm_irq_routing_entry *new_entry)
1029 struct kvm_irq_routing_entry *entry;
1030 int n;
1032 for (n = 0; n < s->irq_routes->nr; n++) {
1033 entry = &s->irq_routes->entries[n];
1034 if (entry->gsi != new_entry->gsi) {
1035 continue;
1038 if(!memcmp(entry, new_entry, sizeof *entry)) {
1039 return 0;
1042 *entry = *new_entry;
1044 kvm_irqchip_commit_routes(s);
1046 return 0;
1049 return -ESRCH;
1052 void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin)
1054 struct kvm_irq_routing_entry e = {};
1056 assert(pin < s->gsi_count);
1058 e.gsi = irq;
1059 e.type = KVM_IRQ_ROUTING_IRQCHIP;
1060 e.flags = 0;
1061 e.u.irqchip.irqchip = irqchip;
1062 e.u.irqchip.pin = pin;
1063 kvm_add_routing_entry(s, &e);
1066 void kvm_irqchip_release_virq(KVMState *s, int virq)
1068 struct kvm_irq_routing_entry *e;
1069 int i;
1071 if (kvm_gsi_direct_mapping()) {
1072 return;
1075 for (i = 0; i < s->irq_routes->nr; i++) {
1076 e = &s->irq_routes->entries[i];
1077 if (e->gsi == virq) {
1078 s->irq_routes->nr--;
1079 *e = s->irq_routes->entries[s->irq_routes->nr];
1082 clear_gsi(s, virq);
1085 static unsigned int kvm_hash_msi(uint32_t data)
1087 /* This is optimized for IA32 MSI layout. However, no other arch shall
1088 * repeat the mistake of not providing a direct MSI injection API. */
1089 return data & 0xff;
1092 static void kvm_flush_dynamic_msi_routes(KVMState *s)
1094 KVMMSIRoute *route, *next;
1095 unsigned int hash;
1097 for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) {
1098 QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) {
1099 kvm_irqchip_release_virq(s, route->kroute.gsi);
1100 QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry);
1101 g_free(route);
1106 static int kvm_irqchip_get_virq(KVMState *s)
1108 uint32_t *word = s->used_gsi_bitmap;
1109 int max_words = ALIGN(s->gsi_count, 32) / 32;
1110 int i, zeroes;
1113 * PIC and IOAPIC share the first 16 GSI numbers, thus the available
1114 * GSI numbers are more than the number of IRQ route. Allocating a GSI
1115 * number can succeed even though a new route entry cannot be added.
1116 * When this happens, flush dynamic MSI entries to free IRQ route entries.
1118 if (!kvm_direct_msi_allowed && s->irq_routes->nr == s->gsi_count) {
1119 kvm_flush_dynamic_msi_routes(s);
1122 /* Return the lowest unused GSI in the bitmap */
1123 for (i = 0; i < max_words; i++) {
1124 zeroes = ctz32(~word[i]);
1125 if (zeroes == 32) {
1126 continue;
1129 return zeroes + i * 32;
1131 return -ENOSPC;
1135 static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg)
1137 unsigned int hash = kvm_hash_msi(msg.data);
1138 KVMMSIRoute *route;
1140 QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) {
1141 if (route->kroute.u.msi.address_lo == (uint32_t)msg.address &&
1142 route->kroute.u.msi.address_hi == (msg.address >> 32) &&
1143 route->kroute.u.msi.data == le32_to_cpu(msg.data)) {
1144 return route;
1147 return NULL;
1150 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1152 struct kvm_msi msi;
1153 KVMMSIRoute *route;
1155 if (kvm_direct_msi_allowed) {
1156 msi.address_lo = (uint32_t)msg.address;
1157 msi.address_hi = msg.address >> 32;
1158 msi.data = le32_to_cpu(msg.data);
1159 msi.flags = 0;
1160 memset(msi.pad, 0, sizeof(msi.pad));
1162 return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi);
1165 route = kvm_lookup_msi_route(s, msg);
1166 if (!route) {
1167 int virq;
1169 virq = kvm_irqchip_get_virq(s);
1170 if (virq < 0) {
1171 return virq;
1174 route = g_malloc0(sizeof(KVMMSIRoute));
1175 route->kroute.gsi = virq;
1176 route->kroute.type = KVM_IRQ_ROUTING_MSI;
1177 route->kroute.flags = 0;
1178 route->kroute.u.msi.address_lo = (uint32_t)msg.address;
1179 route->kroute.u.msi.address_hi = msg.address >> 32;
1180 route->kroute.u.msi.data = le32_to_cpu(msg.data);
1182 kvm_add_routing_entry(s, &route->kroute);
1183 kvm_irqchip_commit_routes(s);
1185 QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route,
1186 entry);
1189 assert(route->kroute.type == KVM_IRQ_ROUTING_MSI);
1191 return kvm_set_irq(s, route->kroute.gsi, 1);
1194 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg, PCIDevice *dev)
1196 struct kvm_irq_routing_entry kroute = {};
1197 int virq;
1199 if (kvm_gsi_direct_mapping()) {
1200 return kvm_arch_msi_data_to_gsi(msg.data);
1203 if (!kvm_gsi_routing_enabled()) {
1204 return -ENOSYS;
1207 virq = kvm_irqchip_get_virq(s);
1208 if (virq < 0) {
1209 return virq;
1212 kroute.gsi = virq;
1213 kroute.type = KVM_IRQ_ROUTING_MSI;
1214 kroute.flags = 0;
1215 kroute.u.msi.address_lo = (uint32_t)msg.address;
1216 kroute.u.msi.address_hi = msg.address >> 32;
1217 kroute.u.msi.data = le32_to_cpu(msg.data);
1218 if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data, dev)) {
1219 kvm_irqchip_release_virq(s, virq);
1220 return -EINVAL;
1223 kvm_add_routing_entry(s, &kroute);
1224 kvm_irqchip_commit_routes(s);
1226 return virq;
1229 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg,
1230 PCIDevice *dev)
1232 struct kvm_irq_routing_entry kroute = {};
1234 if (kvm_gsi_direct_mapping()) {
1235 return 0;
1238 if (!kvm_irqchip_in_kernel()) {
1239 return -ENOSYS;
1242 kroute.gsi = virq;
1243 kroute.type = KVM_IRQ_ROUTING_MSI;
1244 kroute.flags = 0;
1245 kroute.u.msi.address_lo = (uint32_t)msg.address;
1246 kroute.u.msi.address_hi = msg.address >> 32;
1247 kroute.u.msi.data = le32_to_cpu(msg.data);
1248 if (kvm_arch_fixup_msi_route(&kroute, msg.address, msg.data, dev)) {
1249 return -EINVAL;
1252 return kvm_update_routing_entry(s, &kroute);
1255 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int rfd, int virq,
1256 bool assign)
1258 struct kvm_irqfd irqfd = {
1259 .fd = fd,
1260 .gsi = virq,
1261 .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN,
1264 if (rfd != -1) {
1265 irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE;
1266 irqfd.resamplefd = rfd;
1269 if (!kvm_irqfds_enabled()) {
1270 return -ENOSYS;
1273 return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd);
1276 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter)
1278 struct kvm_irq_routing_entry kroute = {};
1279 int virq;
1281 if (!kvm_gsi_routing_enabled()) {
1282 return -ENOSYS;
1285 virq = kvm_irqchip_get_virq(s);
1286 if (virq < 0) {
1287 return virq;
1290 kroute.gsi = virq;
1291 kroute.type = KVM_IRQ_ROUTING_S390_ADAPTER;
1292 kroute.flags = 0;
1293 kroute.u.adapter.summary_addr = adapter->summary_addr;
1294 kroute.u.adapter.ind_addr = adapter->ind_addr;
1295 kroute.u.adapter.summary_offset = adapter->summary_offset;
1296 kroute.u.adapter.ind_offset = adapter->ind_offset;
1297 kroute.u.adapter.adapter_id = adapter->adapter_id;
1299 kvm_add_routing_entry(s, &kroute);
1301 return virq;
1304 int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint)
1306 struct kvm_irq_routing_entry kroute = {};
1307 int virq;
1309 if (!kvm_gsi_routing_enabled()) {
1310 return -ENOSYS;
1312 if (!kvm_check_extension(s, KVM_CAP_HYPERV_SYNIC)) {
1313 return -ENOSYS;
1315 virq = kvm_irqchip_get_virq(s);
1316 if (virq < 0) {
1317 return virq;
1320 kroute.gsi = virq;
1321 kroute.type = KVM_IRQ_ROUTING_HV_SINT;
1322 kroute.flags = 0;
1323 kroute.u.hv_sint.vcpu = vcpu;
1324 kroute.u.hv_sint.sint = sint;
1326 kvm_add_routing_entry(s, &kroute);
1327 kvm_irqchip_commit_routes(s);
1329 return virq;
1332 #else /* !KVM_CAP_IRQ_ROUTING */
1334 void kvm_init_irq_routing(KVMState *s)
1338 void kvm_irqchip_release_virq(KVMState *s, int virq)
1342 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1344 abort();
1347 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1349 return -ENOSYS;
1352 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter)
1354 return -ENOSYS;
1357 int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint)
1359 return -ENOSYS;
1362 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign)
1364 abort();
1367 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1369 return -ENOSYS;
1371 #endif /* !KVM_CAP_IRQ_ROUTING */
1373 int kvm_irqchip_add_irqfd_notifier_gsi(KVMState *s, EventNotifier *n,
1374 EventNotifier *rn, int virq)
1376 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n),
1377 rn ? event_notifier_get_fd(rn) : -1, virq, true);
1380 int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState *s, EventNotifier *n,
1381 int virq)
1383 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), -1, virq,
1384 false);
1387 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n,
1388 EventNotifier *rn, qemu_irq irq)
1390 gpointer key, gsi;
1391 gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi);
1393 if (!found) {
1394 return -ENXIO;
1396 return kvm_irqchip_add_irqfd_notifier_gsi(s, n, rn, GPOINTER_TO_INT(gsi));
1399 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n,
1400 qemu_irq irq)
1402 gpointer key, gsi;
1403 gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi);
1405 if (!found) {
1406 return -ENXIO;
1408 return kvm_irqchip_remove_irqfd_notifier_gsi(s, n, GPOINTER_TO_INT(gsi));
1411 void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi)
1413 g_hash_table_insert(s->gsimap, irq, GINT_TO_POINTER(gsi));
1416 static void kvm_irqchip_create(MachineState *machine, KVMState *s)
1418 int ret;
1420 if (kvm_check_extension(s, KVM_CAP_IRQCHIP)) {
1422 } else if (kvm_check_extension(s, KVM_CAP_S390_IRQCHIP)) {
1423 ret = kvm_vm_enable_cap(s, KVM_CAP_S390_IRQCHIP, 0);
1424 if (ret < 0) {
1425 fprintf(stderr, "Enable kernel irqchip failed: %s\n", strerror(-ret));
1426 exit(1);
1428 } else {
1429 return;
1432 /* First probe and see if there's a arch-specific hook to create the
1433 * in-kernel irqchip for us */
1434 ret = kvm_arch_irqchip_create(machine, s);
1435 if (ret == 0) {
1436 if (machine_kernel_irqchip_split(machine)) {
1437 perror("Split IRQ chip mode not supported.");
1438 exit(1);
1439 } else {
1440 ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP);
1443 if (ret < 0) {
1444 fprintf(stderr, "Create kernel irqchip failed: %s\n", strerror(-ret));
1445 exit(1);
1448 kvm_kernel_irqchip = true;
1449 /* If we have an in-kernel IRQ chip then we must have asynchronous
1450 * interrupt delivery (though the reverse is not necessarily true)
1452 kvm_async_interrupts_allowed = true;
1453 kvm_halt_in_kernel_allowed = true;
1455 kvm_init_irq_routing(s);
1457 s->gsimap = g_hash_table_new(g_direct_hash, g_direct_equal);
1460 /* Find number of supported CPUs using the recommended
1461 * procedure from the kernel API documentation to cope with
1462 * older kernels that may be missing capabilities.
1464 static int kvm_recommended_vcpus(KVMState *s)
1466 int ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS);
1467 return (ret) ? ret : 4;
1470 static int kvm_max_vcpus(KVMState *s)
1472 int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS);
1473 return (ret) ? ret : kvm_recommended_vcpus(s);
1476 static int kvm_init(MachineState *ms)
1478 MachineClass *mc = MACHINE_GET_CLASS(ms);
1479 static const char upgrade_note[] =
1480 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1481 "(see http://sourceforge.net/projects/kvm).\n";
1482 struct {
1483 const char *name;
1484 int num;
1485 } num_cpus[] = {
1486 { "SMP", smp_cpus },
1487 { "hotpluggable", max_cpus },
1488 { NULL, }
1489 }, *nc = num_cpus;
1490 int soft_vcpus_limit, hard_vcpus_limit;
1491 KVMState *s;
1492 const KVMCapabilityInfo *missing_cap;
1493 int ret;
1494 int type = 0;
1495 const char *kvm_type;
1497 s = KVM_STATE(ms->accelerator);
1500 * On systems where the kernel can support different base page
1501 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1502 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1503 * page size for the system though.
1505 assert(TARGET_PAGE_SIZE <= getpagesize());
1507 s->sigmask_len = 8;
1509 #ifdef KVM_CAP_SET_GUEST_DEBUG
1510 QTAILQ_INIT(&s->kvm_sw_breakpoints);
1511 #endif
1512 s->vmfd = -1;
1513 s->fd = qemu_open("/dev/kvm", O_RDWR);
1514 if (s->fd == -1) {
1515 fprintf(stderr, "Could not access KVM kernel module: %m\n");
1516 ret = -errno;
1517 goto err;
1520 ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
1521 if (ret < KVM_API_VERSION) {
1522 if (ret >= 0) {
1523 ret = -EINVAL;
1525 fprintf(stderr, "kvm version too old\n");
1526 goto err;
1529 if (ret > KVM_API_VERSION) {
1530 ret = -EINVAL;
1531 fprintf(stderr, "kvm version not supported\n");
1532 goto err;
1535 s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
1537 /* If unspecified, use the default value */
1538 if (!s->nr_slots) {
1539 s->nr_slots = 32;
1542 /* check the vcpu limits */
1543 soft_vcpus_limit = kvm_recommended_vcpus(s);
1544 hard_vcpus_limit = kvm_max_vcpus(s);
1546 while (nc->name) {
1547 if (nc->num > soft_vcpus_limit) {
1548 fprintf(stderr,
1549 "Warning: Number of %s cpus requested (%d) exceeds "
1550 "the recommended cpus supported by KVM (%d)\n",
1551 nc->name, nc->num, soft_vcpus_limit);
1553 if (nc->num > hard_vcpus_limit) {
1554 fprintf(stderr, "Number of %s cpus requested (%d) exceeds "
1555 "the maximum cpus supported by KVM (%d)\n",
1556 nc->name, nc->num, hard_vcpus_limit);
1557 exit(1);
1560 nc++;
1563 kvm_type = qemu_opt_get(qemu_get_machine_opts(), "kvm-type");
1564 if (mc->kvm_type) {
1565 type = mc->kvm_type(kvm_type);
1566 } else if (kvm_type) {
1567 ret = -EINVAL;
1568 fprintf(stderr, "Invalid argument kvm-type=%s\n", kvm_type);
1569 goto err;
1572 do {
1573 ret = kvm_ioctl(s, KVM_CREATE_VM, type);
1574 } while (ret == -EINTR);
1576 if (ret < 0) {
1577 fprintf(stderr, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret,
1578 strerror(-ret));
1580 #ifdef TARGET_S390X
1581 if (ret == -EINVAL) {
1582 fprintf(stderr,
1583 "Host kernel setup problem detected. Please verify:\n");
1584 fprintf(stderr, "- for kernels supporting the switch_amode or"
1585 " user_mode parameters, whether\n");
1586 fprintf(stderr,
1587 " user space is running in primary address space\n");
1588 fprintf(stderr,
1589 "- for kernels supporting the vm.allocate_pgste sysctl, "
1590 "whether it is enabled\n");
1592 #endif
1593 goto err;
1596 s->vmfd = ret;
1597 missing_cap = kvm_check_extension_list(s, kvm_required_capabilites);
1598 if (!missing_cap) {
1599 missing_cap =
1600 kvm_check_extension_list(s, kvm_arch_required_capabilities);
1602 if (missing_cap) {
1603 ret = -EINVAL;
1604 fprintf(stderr, "kvm does not support %s\n%s",
1605 missing_cap->name, upgrade_note);
1606 goto err;
1609 s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
1611 s->broken_set_mem_region = 1;
1612 ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
1613 if (ret > 0) {
1614 s->broken_set_mem_region = 0;
1617 #ifdef KVM_CAP_VCPU_EVENTS
1618 s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS);
1619 #endif
1621 s->robust_singlestep =
1622 kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP);
1624 #ifdef KVM_CAP_DEBUGREGS
1625 s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS);
1626 #endif
1628 #ifdef KVM_CAP_IRQ_ROUTING
1629 kvm_direct_msi_allowed = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0);
1630 #endif
1632 s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3);
1634 s->irq_set_ioctl = KVM_IRQ_LINE;
1635 if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) {
1636 s->irq_set_ioctl = KVM_IRQ_LINE_STATUS;
1639 #ifdef KVM_CAP_READONLY_MEM
1640 kvm_readonly_mem_allowed =
1641 (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0);
1642 #endif
1644 kvm_eventfds_allowed =
1645 (kvm_check_extension(s, KVM_CAP_IOEVENTFD) > 0);
1647 kvm_irqfds_allowed =
1648 (kvm_check_extension(s, KVM_CAP_IRQFD) > 0);
1650 kvm_resamplefds_allowed =
1651 (kvm_check_extension(s, KVM_CAP_IRQFD_RESAMPLE) > 0);
1653 kvm_vm_attributes_allowed =
1654 (kvm_check_extension(s, KVM_CAP_VM_ATTRIBUTES) > 0);
1656 kvm_ioeventfd_any_length_allowed =
1657 (kvm_check_extension(s, KVM_CAP_IOEVENTFD_ANY_LENGTH) > 0);
1659 ret = kvm_arch_init(ms, s);
1660 if (ret < 0) {
1661 goto err;
1664 if (machine_kernel_irqchip_allowed(ms)) {
1665 kvm_irqchip_create(ms, s);
1668 kvm_state = s;
1670 if (kvm_eventfds_allowed) {
1671 s->memory_listener.listener.eventfd_add = kvm_mem_ioeventfd_add;
1672 s->memory_listener.listener.eventfd_del = kvm_mem_ioeventfd_del;
1674 s->memory_listener.listener.coalesced_mmio_add = kvm_coalesce_mmio_region;
1675 s->memory_listener.listener.coalesced_mmio_del = kvm_uncoalesce_mmio_region;
1677 kvm_memory_listener_register(s, &s->memory_listener,
1678 &address_space_memory, 0);
1679 memory_listener_register(&kvm_io_listener,
1680 &address_space_io);
1682 s->many_ioeventfds = kvm_check_many_ioeventfds();
1684 cpu_interrupt_handler = kvm_handle_interrupt;
1686 return 0;
1688 err:
1689 assert(ret < 0);
1690 if (s->vmfd >= 0) {
1691 close(s->vmfd);
1693 if (s->fd != -1) {
1694 close(s->fd);
1696 g_free(s->memory_listener.slots);
1698 return ret;
1701 void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len)
1703 s->sigmask_len = sigmask_len;
1706 static void kvm_handle_io(uint16_t port, MemTxAttrs attrs, void *data, int direction,
1707 int size, uint32_t count)
1709 int i;
1710 uint8_t *ptr = data;
1712 for (i = 0; i < count; i++) {
1713 address_space_rw(&address_space_io, port, attrs,
1714 ptr, size,
1715 direction == KVM_EXIT_IO_OUT);
1716 ptr += size;
1720 static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run)
1722 fprintf(stderr, "KVM internal error. Suberror: %d\n",
1723 run->internal.suberror);
1725 if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) {
1726 int i;
1728 for (i = 0; i < run->internal.ndata; ++i) {
1729 fprintf(stderr, "extra data[%d]: %"PRIx64"\n",
1730 i, (uint64_t)run->internal.data[i]);
1733 if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) {
1734 fprintf(stderr, "emulation failure\n");
1735 if (!kvm_arch_stop_on_emulation_error(cpu)) {
1736 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1737 return EXCP_INTERRUPT;
1740 /* FIXME: Should trigger a qmp message to let management know
1741 * something went wrong.
1743 return -1;
1746 void kvm_flush_coalesced_mmio_buffer(void)
1748 KVMState *s = kvm_state;
1750 if (s->coalesced_flush_in_progress) {
1751 return;
1754 s->coalesced_flush_in_progress = true;
1756 if (s->coalesced_mmio_ring) {
1757 struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring;
1758 while (ring->first != ring->last) {
1759 struct kvm_coalesced_mmio *ent;
1761 ent = &ring->coalesced_mmio[ring->first];
1763 cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
1764 smp_wmb();
1765 ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
1769 s->coalesced_flush_in_progress = false;
1772 static void do_kvm_cpu_synchronize_state(void *arg)
1774 CPUState *cpu = arg;
1776 if (!cpu->kvm_vcpu_dirty) {
1777 kvm_arch_get_registers(cpu);
1778 cpu->kvm_vcpu_dirty = true;
1782 void kvm_cpu_synchronize_state(CPUState *cpu)
1784 if (!cpu->kvm_vcpu_dirty) {
1785 run_on_cpu(cpu, do_kvm_cpu_synchronize_state, cpu);
1789 static void do_kvm_cpu_synchronize_post_reset(void *arg)
1791 CPUState *cpu = arg;
1793 kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE);
1794 cpu->kvm_vcpu_dirty = false;
1797 void kvm_cpu_synchronize_post_reset(CPUState *cpu)
1799 run_on_cpu(cpu, do_kvm_cpu_synchronize_post_reset, cpu);
1802 static void do_kvm_cpu_synchronize_post_init(void *arg)
1804 CPUState *cpu = arg;
1806 kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE);
1807 cpu->kvm_vcpu_dirty = false;
1810 void kvm_cpu_synchronize_post_init(CPUState *cpu)
1812 run_on_cpu(cpu, do_kvm_cpu_synchronize_post_init, cpu);
1815 int kvm_cpu_exec(CPUState *cpu)
1817 struct kvm_run *run = cpu->kvm_run;
1818 int ret, run_ret;
1820 DPRINTF("kvm_cpu_exec()\n");
1822 if (kvm_arch_process_async_events(cpu)) {
1823 cpu->exit_request = 0;
1824 return EXCP_HLT;
1827 qemu_mutex_unlock_iothread();
1829 do {
1830 MemTxAttrs attrs;
1832 if (cpu->kvm_vcpu_dirty) {
1833 kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE);
1834 cpu->kvm_vcpu_dirty = false;
1837 kvm_arch_pre_run(cpu, run);
1838 if (cpu->exit_request) {
1839 DPRINTF("interrupt exit requested\n");
1841 * KVM requires us to reenter the kernel after IO exits to complete
1842 * instruction emulation. This self-signal will ensure that we
1843 * leave ASAP again.
1845 qemu_cpu_kick_self();
1848 run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0);
1850 attrs = kvm_arch_post_run(cpu, run);
1852 if (run_ret < 0) {
1853 if (run_ret == -EINTR || run_ret == -EAGAIN) {
1854 DPRINTF("io window exit\n");
1855 ret = EXCP_INTERRUPT;
1856 break;
1858 fprintf(stderr, "error: kvm run failed %s\n",
1859 strerror(-run_ret));
1860 #ifdef TARGET_PPC
1861 if (run_ret == -EBUSY) {
1862 fprintf(stderr,
1863 "This is probably because your SMT is enabled.\n"
1864 "VCPU can only run on primary threads with all "
1865 "secondary threads offline.\n");
1867 #endif
1868 ret = -1;
1869 break;
1872 trace_kvm_run_exit(cpu->cpu_index, run->exit_reason);
1873 switch (run->exit_reason) {
1874 case KVM_EXIT_IO:
1875 DPRINTF("handle_io\n");
1876 /* Called outside BQL */
1877 kvm_handle_io(run->io.port, attrs,
1878 (uint8_t *)run + run->io.data_offset,
1879 run->io.direction,
1880 run->io.size,
1881 run->io.count);
1882 ret = 0;
1883 break;
1884 case KVM_EXIT_MMIO:
1885 DPRINTF("handle_mmio\n");
1886 /* Called outside BQL */
1887 address_space_rw(&address_space_memory,
1888 run->mmio.phys_addr, attrs,
1889 run->mmio.data,
1890 run->mmio.len,
1891 run->mmio.is_write);
1892 ret = 0;
1893 break;
1894 case KVM_EXIT_IRQ_WINDOW_OPEN:
1895 DPRINTF("irq_window_open\n");
1896 ret = EXCP_INTERRUPT;
1897 break;
1898 case KVM_EXIT_SHUTDOWN:
1899 DPRINTF("shutdown\n");
1900 qemu_system_reset_request();
1901 ret = EXCP_INTERRUPT;
1902 break;
1903 case KVM_EXIT_UNKNOWN:
1904 fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n",
1905 (uint64_t)run->hw.hardware_exit_reason);
1906 ret = -1;
1907 break;
1908 case KVM_EXIT_INTERNAL_ERROR:
1909 ret = kvm_handle_internal_error(cpu, run);
1910 break;
1911 case KVM_EXIT_SYSTEM_EVENT:
1912 switch (run->system_event.type) {
1913 case KVM_SYSTEM_EVENT_SHUTDOWN:
1914 qemu_system_shutdown_request();
1915 ret = EXCP_INTERRUPT;
1916 break;
1917 case KVM_SYSTEM_EVENT_RESET:
1918 qemu_system_reset_request();
1919 ret = EXCP_INTERRUPT;
1920 break;
1921 case KVM_SYSTEM_EVENT_CRASH:
1922 qemu_mutex_lock_iothread();
1923 qemu_system_guest_panicked();
1924 qemu_mutex_unlock_iothread();
1925 ret = 0;
1926 break;
1927 default:
1928 DPRINTF("kvm_arch_handle_exit\n");
1929 ret = kvm_arch_handle_exit(cpu, run);
1930 break;
1932 break;
1933 default:
1934 DPRINTF("kvm_arch_handle_exit\n");
1935 ret = kvm_arch_handle_exit(cpu, run);
1936 break;
1938 } while (ret == 0);
1940 qemu_mutex_lock_iothread();
1942 if (ret < 0) {
1943 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1944 vm_stop(RUN_STATE_INTERNAL_ERROR);
1947 cpu->exit_request = 0;
1948 return ret;
1951 int kvm_ioctl(KVMState *s, int type, ...)
1953 int ret;
1954 void *arg;
1955 va_list ap;
1957 va_start(ap, type);
1958 arg = va_arg(ap, void *);
1959 va_end(ap);
1961 trace_kvm_ioctl(type, arg);
1962 ret = ioctl(s->fd, type, arg);
1963 if (ret == -1) {
1964 ret = -errno;
1966 return ret;
1969 int kvm_vm_ioctl(KVMState *s, int type, ...)
1971 int ret;
1972 void *arg;
1973 va_list ap;
1975 va_start(ap, type);
1976 arg = va_arg(ap, void *);
1977 va_end(ap);
1979 trace_kvm_vm_ioctl(type, arg);
1980 ret = ioctl(s->vmfd, type, arg);
1981 if (ret == -1) {
1982 ret = -errno;
1984 return ret;
1987 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...)
1989 int ret;
1990 void *arg;
1991 va_list ap;
1993 va_start(ap, type);
1994 arg = va_arg(ap, void *);
1995 va_end(ap);
1997 trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg);
1998 ret = ioctl(cpu->kvm_fd, type, arg);
1999 if (ret == -1) {
2000 ret = -errno;
2002 return ret;
2005 int kvm_device_ioctl(int fd, int type, ...)
2007 int ret;
2008 void *arg;
2009 va_list ap;
2011 va_start(ap, type);
2012 arg = va_arg(ap, void *);
2013 va_end(ap);
2015 trace_kvm_device_ioctl(fd, type, arg);
2016 ret = ioctl(fd, type, arg);
2017 if (ret == -1) {
2018 ret = -errno;
2020 return ret;
2023 int kvm_vm_check_attr(KVMState *s, uint32_t group, uint64_t attr)
2025 int ret;
2026 struct kvm_device_attr attribute = {
2027 .group = group,
2028 .attr = attr,
2031 if (!kvm_vm_attributes_allowed) {
2032 return 0;
2035 ret = kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attribute);
2036 /* kvm returns 0 on success for HAS_DEVICE_ATTR */
2037 return ret ? 0 : 1;
2040 int kvm_device_check_attr(int dev_fd, uint32_t group, uint64_t attr)
2042 struct kvm_device_attr attribute = {
2043 .group = group,
2044 .attr = attr,
2045 .flags = 0,
2048 return kvm_device_ioctl(dev_fd, KVM_HAS_DEVICE_ATTR, &attribute) ? 0 : 1;
2051 void kvm_device_access(int fd, int group, uint64_t attr,
2052 void *val, bool write)
2054 struct kvm_device_attr kvmattr;
2055 int err;
2057 kvmattr.flags = 0;
2058 kvmattr.group = group;
2059 kvmattr.attr = attr;
2060 kvmattr.addr = (uintptr_t)val;
2062 err = kvm_device_ioctl(fd,
2063 write ? KVM_SET_DEVICE_ATTR : KVM_GET_DEVICE_ATTR,
2064 &kvmattr);
2065 if (err < 0) {
2066 error_report("KVM_%s_DEVICE_ATTR failed: %s",
2067 write ? "SET" : "GET", strerror(-err));
2068 error_printf("Group %d attr 0x%016" PRIx64, group, attr);
2069 abort();
2073 int kvm_has_sync_mmu(void)
2075 return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU);
2078 int kvm_has_vcpu_events(void)
2080 return kvm_state->vcpu_events;
2083 int kvm_has_robust_singlestep(void)
2085 return kvm_state->robust_singlestep;
2088 int kvm_has_debugregs(void)
2090 return kvm_state->debugregs;
2093 int kvm_has_many_ioeventfds(void)
2095 if (!kvm_enabled()) {
2096 return 0;
2098 return kvm_state->many_ioeventfds;
2101 int kvm_has_gsi_routing(void)
2103 #ifdef KVM_CAP_IRQ_ROUTING
2104 return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING);
2105 #else
2106 return false;
2107 #endif
2110 int kvm_has_intx_set_mask(void)
2112 return kvm_state->intx_set_mask;
2115 void kvm_setup_guest_memory(void *start, size_t size)
2117 if (!kvm_has_sync_mmu()) {
2118 int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK);
2120 if (ret) {
2121 perror("qemu_madvise");
2122 fprintf(stderr,
2123 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
2124 exit(1);
2129 #ifdef KVM_CAP_SET_GUEST_DEBUG
2130 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu,
2131 target_ulong pc)
2133 struct kvm_sw_breakpoint *bp;
2135 QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) {
2136 if (bp->pc == pc) {
2137 return bp;
2140 return NULL;
2143 int kvm_sw_breakpoints_active(CPUState *cpu)
2145 return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints);
2148 struct kvm_set_guest_debug_data {
2149 struct kvm_guest_debug dbg;
2150 CPUState *cpu;
2151 int err;
2154 static void kvm_invoke_set_guest_debug(void *data)
2156 struct kvm_set_guest_debug_data *dbg_data = data;
2158 dbg_data->err = kvm_vcpu_ioctl(dbg_data->cpu, KVM_SET_GUEST_DEBUG,
2159 &dbg_data->dbg);
2162 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2164 struct kvm_set_guest_debug_data data;
2166 data.dbg.control = reinject_trap;
2168 if (cpu->singlestep_enabled) {
2169 data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
2171 kvm_arch_update_guest_debug(cpu, &data.dbg);
2172 data.cpu = cpu;
2174 run_on_cpu(cpu, kvm_invoke_set_guest_debug, &data);
2175 return data.err;
2178 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2179 target_ulong len, int type)
2181 struct kvm_sw_breakpoint *bp;
2182 int err;
2184 if (type == GDB_BREAKPOINT_SW) {
2185 bp = kvm_find_sw_breakpoint(cpu, addr);
2186 if (bp) {
2187 bp->use_count++;
2188 return 0;
2191 bp = g_malloc(sizeof(struct kvm_sw_breakpoint));
2192 bp->pc = addr;
2193 bp->use_count = 1;
2194 err = kvm_arch_insert_sw_breakpoint(cpu, bp);
2195 if (err) {
2196 g_free(bp);
2197 return err;
2200 QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
2201 } else {
2202 err = kvm_arch_insert_hw_breakpoint(addr, len, type);
2203 if (err) {
2204 return err;
2208 CPU_FOREACH(cpu) {
2209 err = kvm_update_guest_debug(cpu, 0);
2210 if (err) {
2211 return err;
2214 return 0;
2217 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2218 target_ulong len, int type)
2220 struct kvm_sw_breakpoint *bp;
2221 int err;
2223 if (type == GDB_BREAKPOINT_SW) {
2224 bp = kvm_find_sw_breakpoint(cpu, addr);
2225 if (!bp) {
2226 return -ENOENT;
2229 if (bp->use_count > 1) {
2230 bp->use_count--;
2231 return 0;
2234 err = kvm_arch_remove_sw_breakpoint(cpu, bp);
2235 if (err) {
2236 return err;
2239 QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
2240 g_free(bp);
2241 } else {
2242 err = kvm_arch_remove_hw_breakpoint(addr, len, type);
2243 if (err) {
2244 return err;
2248 CPU_FOREACH(cpu) {
2249 err = kvm_update_guest_debug(cpu, 0);
2250 if (err) {
2251 return err;
2254 return 0;
2257 void kvm_remove_all_breakpoints(CPUState *cpu)
2259 struct kvm_sw_breakpoint *bp, *next;
2260 KVMState *s = cpu->kvm_state;
2261 CPUState *tmpcpu;
2263 QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
2264 if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) {
2265 /* Try harder to find a CPU that currently sees the breakpoint. */
2266 CPU_FOREACH(tmpcpu) {
2267 if (kvm_arch_remove_sw_breakpoint(tmpcpu, bp) == 0) {
2268 break;
2272 QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry);
2273 g_free(bp);
2275 kvm_arch_remove_all_hw_breakpoints();
2277 CPU_FOREACH(cpu) {
2278 kvm_update_guest_debug(cpu, 0);
2282 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2284 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2286 return -EINVAL;
2289 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2290 target_ulong len, int type)
2292 return -EINVAL;
2295 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2296 target_ulong len, int type)
2298 return -EINVAL;
2301 void kvm_remove_all_breakpoints(CPUState *cpu)
2304 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2306 int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset)
2308 KVMState *s = kvm_state;
2309 struct kvm_signal_mask *sigmask;
2310 int r;
2312 if (!sigset) {
2313 return kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, NULL);
2316 sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset));
2318 sigmask->len = s->sigmask_len;
2319 memcpy(sigmask->sigset, sigset, sizeof(*sigset));
2320 r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask);
2321 g_free(sigmask);
2323 return r;
2325 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2327 return kvm_arch_on_sigbus_vcpu(cpu, code, addr);
2330 int kvm_on_sigbus(int code, void *addr)
2332 return kvm_arch_on_sigbus(code, addr);
2335 int kvm_create_device(KVMState *s, uint64_t type, bool test)
2337 int ret;
2338 struct kvm_create_device create_dev;
2340 create_dev.type = type;
2341 create_dev.fd = -1;
2342 create_dev.flags = test ? KVM_CREATE_DEVICE_TEST : 0;
2344 if (!kvm_check_extension(s, KVM_CAP_DEVICE_CTRL)) {
2345 return -ENOTSUP;
2348 ret = kvm_vm_ioctl(s, KVM_CREATE_DEVICE, &create_dev);
2349 if (ret) {
2350 return ret;
2353 return test ? 0 : create_dev.fd;
2356 int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source)
2358 struct kvm_one_reg reg;
2359 int r;
2361 reg.id = id;
2362 reg.addr = (uintptr_t) source;
2363 r = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
2364 if (r) {
2365 trace_kvm_failed_reg_set(id, strerror(r));
2367 return r;
2370 int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target)
2372 struct kvm_one_reg reg;
2373 int r;
2375 reg.id = id;
2376 reg.addr = (uintptr_t) target;
2377 r = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
2378 if (r) {
2379 trace_kvm_failed_reg_get(id, strerror(r));
2381 return r;
2384 static void kvm_accel_class_init(ObjectClass *oc, void *data)
2386 AccelClass *ac = ACCEL_CLASS(oc);
2387 ac->name = "KVM";
2388 ac->init_machine = kvm_init;
2389 ac->allowed = &kvm_allowed;
2392 static const TypeInfo kvm_accel_type = {
2393 .name = TYPE_KVM_ACCEL,
2394 .parent = TYPE_ACCEL,
2395 .class_init = kvm_accel_class_init,
2396 .instance_size = sizeof(KVMState),
2399 static void kvm_type_init(void)
2401 type_register_static(&kvm_accel_type);
2404 type_init(kvm_type_init);