scripts/qmp: Use Python 2.6 "except E as ..." syntax
[qemu/kevin.git] / kvm-all.c
blob9cc9ba6ea6fdfaee58b56e84b7c5fee0068e1546
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 "qemu/osdep.h"
17 #include <sys/ioctl.h>
18 #include <sys/mman.h>
20 #include <linux/kvm.h>
22 #include "qemu-common.h"
23 #include "qemu/atomic.h"
24 #include "qemu/option.h"
25 #include "qemu/config-file.h"
26 #include "qemu/error-report.h"
27 #include "hw/hw.h"
28 #include "hw/pci/msi.h"
29 #include "hw/s390x/adapter.h"
30 #include "exec/gdbstub.h"
31 #include "sysemu/kvm_int.h"
32 #include "qemu/bswap.h"
33 #include "exec/memory.h"
34 #include "exec/ram_addr.h"
35 #include "exec/address-spaces.h"
36 #include "qemu/event_notifier.h"
37 #include "trace.h"
38 #include "hw/irq.h"
40 #include "hw/boards.h"
42 /* This check must be after config-host.h is included */
43 #ifdef CONFIG_EVENTFD
44 #include <sys/eventfd.h>
45 #endif
47 /* KVM uses PAGE_SIZE in its definition of KVM_COALESCED_MMIO_MAX. We
48 * need to use the real host PAGE_SIZE, as that's what KVM will use.
50 #define PAGE_SIZE getpagesize()
52 //#define DEBUG_KVM
54 #ifdef DEBUG_KVM
55 #define DPRINTF(fmt, ...) \
56 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
57 #else
58 #define DPRINTF(fmt, ...) \
59 do { } while (0)
60 #endif
62 #define KVM_MSI_HASHTAB_SIZE 256
64 struct KVMState
66 AccelState parent_obj;
68 int nr_slots;
69 int fd;
70 int vmfd;
71 int coalesced_mmio;
72 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
73 bool coalesced_flush_in_progress;
74 int broken_set_mem_region;
75 int vcpu_events;
76 int robust_singlestep;
77 int debugregs;
78 #ifdef KVM_CAP_SET_GUEST_DEBUG
79 struct kvm_sw_breakpoint_head kvm_sw_breakpoints;
80 #endif
81 int many_ioeventfds;
82 int intx_set_mask;
83 /* The man page (and posix) say ioctl numbers are signed int, but
84 * they're not. Linux, glibc and *BSD all treat ioctl numbers as
85 * unsigned, and treating them as signed here can break things */
86 unsigned irq_set_ioctl;
87 unsigned int sigmask_len;
88 GHashTable *gsimap;
89 #ifdef KVM_CAP_IRQ_ROUTING
90 struct kvm_irq_routing *irq_routes;
91 int nr_allocated_irq_routes;
92 uint32_t *used_gsi_bitmap;
93 unsigned int gsi_count;
94 QTAILQ_HEAD(msi_hashtab, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE];
95 #endif
96 KVMMemoryListener memory_listener;
99 KVMState *kvm_state;
100 bool kvm_kernel_irqchip;
101 bool kvm_split_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;
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
122 static KVMSlot *kvm_get_free_slot(KVMMemoryListener *kml)
124 KVMState *s = kvm_state;
125 int i;
127 for (i = 0; i < s->nr_slots; i++) {
128 if (kml->slots[i].memory_size == 0) {
129 return &kml->slots[i];
133 return NULL;
136 bool kvm_has_free_slot(MachineState *ms)
138 KVMState *s = KVM_STATE(ms->accelerator);
140 return kvm_get_free_slot(&s->memory_listener);
143 static KVMSlot *kvm_alloc_slot(KVMMemoryListener *kml)
145 KVMSlot *slot = kvm_get_free_slot(kml);
147 if (slot) {
148 return slot;
151 fprintf(stderr, "%s: no free slot available\n", __func__);
152 abort();
155 static KVMSlot *kvm_lookup_matching_slot(KVMMemoryListener *kml,
156 hwaddr start_addr,
157 hwaddr end_addr)
159 KVMState *s = kvm_state;
160 int i;
162 for (i = 0; i < s->nr_slots; i++) {
163 KVMSlot *mem = &kml->slots[i];
165 if (start_addr == mem->start_addr &&
166 end_addr == mem->start_addr + mem->memory_size) {
167 return mem;
171 return NULL;
175 * Find overlapping slot with lowest start address
177 static KVMSlot *kvm_lookup_overlapping_slot(KVMMemoryListener *kml,
178 hwaddr start_addr,
179 hwaddr end_addr)
181 KVMState *s = kvm_state;
182 KVMSlot *found = NULL;
183 int i;
185 for (i = 0; i < s->nr_slots; i++) {
186 KVMSlot *mem = &kml->slots[i];
188 if (mem->memory_size == 0 ||
189 (found && found->start_addr < mem->start_addr)) {
190 continue;
193 if (end_addr > mem->start_addr &&
194 start_addr < mem->start_addr + mem->memory_size) {
195 found = mem;
199 return found;
202 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram,
203 hwaddr *phys_addr)
205 KVMMemoryListener *kml = &s->memory_listener;
206 int i;
208 for (i = 0; i < s->nr_slots; i++) {
209 KVMSlot *mem = &kml->slots[i];
211 if (ram >= mem->ram && ram < mem->ram + mem->memory_size) {
212 *phys_addr = mem->start_addr + (ram - mem->ram);
213 return 1;
217 return 0;
220 static int kvm_set_user_memory_region(KVMMemoryListener *kml, KVMSlot *slot)
222 KVMState *s = kvm_state;
223 struct kvm_userspace_memory_region mem;
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;
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);
236 mem.memory_size = slot->memory_size;
237 return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
240 int kvm_init_vcpu(CPUState *cpu)
242 KVMState *s = kvm_state;
243 long mmap_size;
244 int ret;
246 DPRINTF("kvm_init_vcpu\n");
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;
254 cpu->kvm_fd = ret;
255 cpu->kvm_state = s;
256 cpu->kvm_vcpu_dirty = true;
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;
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;
273 if (s->coalesced_mmio && !s->coalesced_mmio_ring) {
274 s->coalesced_mmio_ring =
275 (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE;
278 ret = kvm_arch_init_vcpu(cpu);
279 err:
280 return ret;
284 * dirty pages logging control
287 static int kvm_mem_flags(MemoryRegion *mr)
289 bool readonly = mr->readonly || memory_region_is_romd(mr);
290 int flags = 0;
292 if (memory_region_get_dirty_log_mask(mr) != 0) {
293 flags |= KVM_MEM_LOG_DIRTY_PAGES;
295 if (readonly && kvm_readonly_mem_allowed) {
296 flags |= KVM_MEM_READONLY;
298 return flags;
301 static int kvm_slot_update_flags(KVMMemoryListener *kml, KVMSlot *mem,
302 MemoryRegion *mr)
304 int old_flags;
306 old_flags = mem->flags;
307 mem->flags = kvm_mem_flags(mr);
309 /* If nothing changed effectively, no need to issue ioctl */
310 if (mem->flags == old_flags) {
311 return 0;
314 return kvm_set_user_memory_region(kml, mem);
317 static int kvm_section_update_flags(KVMMemoryListener *kml,
318 MemoryRegionSection *section)
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);
324 if (mem == NULL) {
325 return 0;
326 } else {
327 return kvm_slot_update_flags(kml, mem, section->mr);
331 static void kvm_log_start(MemoryListener *listener,
332 MemoryRegionSection *section,
333 int old, int new)
335 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
336 int r;
338 if (old != 0) {
339 return;
342 r = kvm_section_update_flags(kml, section);
343 if (r < 0) {
344 abort();
348 static void kvm_log_stop(MemoryListener *listener,
349 MemoryRegionSection *section,
350 int old, int new)
352 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
353 int r;
355 if (new != 0) {
356 return;
359 r = kvm_section_update_flags(kml, section);
360 if (r < 0) {
361 abort();
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)
369 ram_addr_t start = section->offset_within_region + section->mr->ram_addr;
370 ram_addr_t pages = int128_get64(section->size) / getpagesize();
372 cpu_physical_memory_set_dirty_lebitmap(bitmap, start, pages);
373 return 0;
376 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
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.
384 * @start_add: start of logged region.
385 * @end_addr: end of logged region.
387 static int kvm_physical_sync_dirty_bitmap(KVMMemoryListener *kml,
388 MemoryRegionSection *section)
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);
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;
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.
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);
424 allocated_size = size;
425 memset(d.dirty_bitmap, 0, allocated_size);
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;
434 kvm_get_dirty_pages_log_range(section, d.dirty_bitmap);
435 start_addr = mem->start_addr + mem->memory_size;
437 g_free(d.dirty_bitmap);
439 return ret;
442 static void kvm_coalesce_mmio_region(MemoryListener *listener,
443 MemoryRegionSection *secion,
444 hwaddr start, hwaddr size)
446 KVMState *s = kvm_state;
448 if (s->coalesced_mmio) {
449 struct kvm_coalesced_mmio_zone zone;
451 zone.addr = start;
452 zone.size = size;
453 zone.pad = 0;
455 (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
459 static void kvm_uncoalesce_mmio_region(MemoryListener *listener,
460 MemoryRegionSection *secion,
461 hwaddr start, hwaddr size)
463 KVMState *s = kvm_state;
465 if (s->coalesced_mmio) {
466 struct kvm_coalesced_mmio_zone zone;
468 zone.addr = start;
469 zone.size = size;
470 zone.pad = 0;
472 (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
476 int kvm_check_extension(KVMState *s, unsigned int extension)
478 int ret;
480 ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
481 if (ret < 0) {
482 ret = 0;
485 return ret;
488 int kvm_vm_check_extension(KVMState *s, unsigned int extension)
490 int ret;
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);
498 return ret;
501 static uint32_t adjust_ioeventfd_endianness(uint32_t val, uint32_t size)
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.
509 switch (size) {
510 case 2:
511 val = bswap16(val);
512 break;
513 case 4:
514 val = bswap32(val);
515 break;
517 #endif
518 return val;
521 static int kvm_set_ioeventfd_mmio(int fd, hwaddr addr, uint32_t val,
522 bool assign, uint32_t size, bool datamatch)
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,
533 if (!kvm_enabled()) {
534 return -ENOSYS;
537 if (datamatch) {
538 iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
540 if (!assign) {
541 iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
544 ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd);
546 if (ret < 0) {
547 return -errno;
550 return 0;
553 static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val,
554 bool assign, uint32_t size, bool datamatch)
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,
563 int r;
564 if (!kvm_enabled()) {
565 return -ENOSYS;
567 if (datamatch) {
568 kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
570 if (!assign) {
571 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
573 r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
574 if (r < 0) {
575 return r;
577 return 0;
581 static int kvm_check_many_ioeventfds(void)
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.
587 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
588 * can avoid creating too many ioeventfds.
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;
598 ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true);
599 if (ret < 0) {
600 close(ioeventfds[i]);
601 break;
605 /* Decide whether many devices are supported or not */
606 ret = i == ARRAY_SIZE(ioeventfds);
608 while (i-- > 0) {
609 kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true);
610 close(ioeventfds[i]);
612 return ret;
613 #else
614 return 0;
615 #endif
618 static const KVMCapabilityInfo *
619 kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list)
621 while (list->name) {
622 if (!kvm_check_extension(s, list->value)) {
623 return list;
625 list++;
627 return NULL;
630 static void kvm_set_phys_mem(KVMMemoryListener *kml,
631 MemoryRegionSection *section, bool add)
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;
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;
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;
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;
668 ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta;
670 while (1) {
671 mem = kvm_lookup_overlapping_slot(kml, start_addr, start_addr + size);
672 if (!mem) {
673 break;
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;
685 old = *mem;
687 if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
688 kvm_physical_sync_dirty_bitmap(kml, section);
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();
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.
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);
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();
723 start_addr += old.memory_size;
724 ram += old.memory_size;
725 size -= old.memory_size;
726 continue;
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);
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();
750 /* register suffix slot */
751 if (old.start_addr + old.memory_size > start_addr + size) {
752 ram_addr_t size_delta;
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);
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();
770 /* in case the KVM bug workaround already "consumed" the new slot */
771 if (!size) {
772 return;
774 if (!add) {
775 return;
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);
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();
791 static void kvm_region_add(MemoryListener *listener,
792 MemoryRegionSection *section)
794 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
796 memory_region_ref(section->mr);
797 kvm_set_phys_mem(kml, section, true);
800 static void kvm_region_del(MemoryListener *listener,
801 MemoryRegionSection *section)
803 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
805 kvm_set_phys_mem(kml, section, false);
806 memory_region_unref(section->mr);
809 static void kvm_log_sync(MemoryListener *listener,
810 MemoryRegionSection *section)
812 KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
813 int r;
815 r = kvm_physical_sync_dirty_bitmap(kml, section);
816 if (r < 0) {
817 abort();
821 static void kvm_mem_ioeventfd_add(MemoryListener *listener,
822 MemoryRegionSection *section,
823 bool match_data, uint64_t data,
824 EventNotifier *e)
826 int fd = event_notifier_get_fd(e);
827 int r;
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();
839 static void kvm_mem_ioeventfd_del(MemoryListener *listener,
840 MemoryRegionSection *section,
841 bool match_data, uint64_t data,
842 EventNotifier *e)
844 int fd = event_notifier_get_fd(e);
845 int r;
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();
855 static void kvm_io_ioeventfd_add(MemoryListener *listener,
856 MemoryRegionSection *section,
857 bool match_data, uint64_t data,
858 EventNotifier *e)
860 int fd = event_notifier_get_fd(e);
861 int r;
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();
873 static void kvm_io_ioeventfd_del(MemoryListener *listener,
874 MemoryRegionSection *section,
875 bool match_data, uint64_t data,
876 EventNotifier *e)
879 int fd = event_notifier_get_fd(e);
880 int r;
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();
890 void kvm_memory_listener_register(KVMState *s, KVMMemoryListener *kml,
891 AddressSpace *as, int as_id)
893 int i;
895 kml->slots = g_malloc0(s->nr_slots * sizeof(KVMSlot));
896 kml->as_id = as_id;
898 for (i = 0; i < s->nr_slots; i++) {
899 kml->slots[i].slot = i;
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;
909 memory_listener_register(&kml->listener, as);
912 static MemoryListener kvm_io_listener = {
913 .eventfd_add = kvm_io_ioeventfd_add,
914 .eventfd_del = kvm_io_ioeventfd_del,
915 .priority = 10,
918 static void kvm_handle_interrupt(CPUState *cpu, int mask)
920 cpu->interrupt_request |= mask;
922 if (!qemu_cpu_is_self(cpu)) {
923 qemu_cpu_kick(cpu);
927 int kvm_set_irq(KVMState *s, int irq, int level)
929 struct kvm_irq_level event;
930 int ret;
932 assert(kvm_async_interrupts_enabled());
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();
942 return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status;
945 #ifdef KVM_CAP_IRQ_ROUTING
946 typedef struct KVMMSIRoute {
947 struct kvm_irq_routing_entry kroute;
948 QTAILQ_ENTRY(KVMMSIRoute) entry;
949 } KVMMSIRoute;
951 static void set_gsi(KVMState *s, unsigned int gsi)
953 s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32);
956 static void clear_gsi(KVMState *s, unsigned int gsi)
958 s->used_gsi_bitmap[gsi / 32] &= ~(1U << (gsi % 32));
961 void kvm_init_irq_routing(KVMState *s)
963 int gsi_count, i;
965 gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING) - 1;
966 if (gsi_count > 0) {
967 unsigned int gsi_bits, i;
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;
974 /* Mark any over-allocated bits as already in use */
975 for (i = gsi_count; i < gsi_bits; i++) {
976 set_gsi(s, i);
980 s->irq_routes = g_malloc0(sizeof(*s->irq_routes));
981 s->nr_allocated_irq_routes = 0;
983 if (!kvm_direct_msi_allowed) {
984 for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) {
985 QTAILQ_INIT(&s->msi_hashtab[i]);
989 kvm_arch_init_irq_routing(s);
992 void kvm_irqchip_commit_routes(KVMState *s)
994 int ret;
996 s->irq_routes->flags = 0;
997 ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes);
998 assert(ret == 0);
1001 static void kvm_add_routing_entry(KVMState *s,
1002 struct kvm_irq_routing_entry *entry)
1004 struct kvm_irq_routing_entry *new;
1005 int n, size;
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;
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;
1017 n = s->irq_routes->nr++;
1018 new = &s->irq_routes->entries[n];
1020 *new = *entry;
1022 set_gsi(s, entry->gsi);
1025 static int kvm_update_routing_entry(KVMState *s,
1026 struct kvm_irq_routing_entry *new_entry)
1028 struct kvm_irq_routing_entry *entry;
1029 int n;
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;
1037 if(!memcmp(entry, new_entry, sizeof *entry)) {
1038 return 0;
1041 *entry = *new_entry;
1043 kvm_irqchip_commit_routes(s);
1045 return 0;
1048 return -ESRCH;
1051 void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin)
1053 struct kvm_irq_routing_entry e = {};
1055 assert(pin < s->gsi_count);
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);
1065 void kvm_irqchip_release_virq(KVMState *s, int virq)
1067 struct kvm_irq_routing_entry *e;
1068 int i;
1070 if (kvm_gsi_direct_mapping()) {
1071 return;
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];
1081 clear_gsi(s, virq);
1084 static unsigned int kvm_hash_msi(uint32_t data)
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;
1091 static void kvm_flush_dynamic_msi_routes(KVMState *s)
1093 KVMMSIRoute *route, *next;
1094 unsigned int hash;
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);
1105 static int kvm_irqchip_get_virq(KVMState *s)
1107 uint32_t *word = s->used_gsi_bitmap;
1108 int max_words = ALIGN(s->gsi_count, 32) / 32;
1109 int i, zeroes;
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.
1117 if (!kvm_direct_msi_allowed && s->irq_routes->nr == s->gsi_count) {
1118 kvm_flush_dynamic_msi_routes(s);
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;
1128 return zeroes + i * 32;
1130 return -ENOSPC;
1134 static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg)
1136 unsigned int hash = kvm_hash_msi(msg.data);
1137 KVMMSIRoute *route;
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;
1146 return NULL;
1149 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1151 struct kvm_msi msi;
1152 KVMMSIRoute *route;
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));
1161 return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi);
1164 route = kvm_lookup_msi_route(s, msg);
1165 if (!route) {
1166 int virq;
1168 virq = kvm_irqchip_get_virq(s);
1169 if (virq < 0) {
1170 return virq;
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);
1181 kvm_add_routing_entry(s, &route->kroute);
1182 kvm_irqchip_commit_routes(s);
1184 QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route,
1185 entry);
1188 assert(route->kroute.type == KVM_IRQ_ROUTING_MSI);
1190 return kvm_set_irq(s, route->kroute.gsi, 1);
1193 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg, PCIDevice *dev)
1195 struct kvm_irq_routing_entry kroute = {};
1196 int virq;
1198 if (kvm_gsi_direct_mapping()) {
1199 return kvm_arch_msi_data_to_gsi(msg.data);
1202 if (!kvm_gsi_routing_enabled()) {
1203 return -ENOSYS;
1206 virq = kvm_irqchip_get_virq(s);
1207 if (virq < 0) {
1208 return virq;
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;
1222 kvm_add_routing_entry(s, &kroute);
1223 kvm_irqchip_commit_routes(s);
1225 return virq;
1228 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg,
1229 PCIDevice *dev)
1231 struct kvm_irq_routing_entry kroute = {};
1233 if (kvm_gsi_direct_mapping()) {
1234 return 0;
1237 if (!kvm_irqchip_in_kernel()) {
1238 return -ENOSYS;
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;
1251 return kvm_update_routing_entry(s, &kroute);
1254 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int rfd, int virq,
1255 bool assign)
1257 struct kvm_irqfd irqfd = {
1258 .fd = fd,
1259 .gsi = virq,
1260 .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN,
1263 if (rfd != -1) {
1264 irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE;
1265 irqfd.resamplefd = rfd;
1268 if (!kvm_irqfds_enabled()) {
1269 return -ENOSYS;
1272 return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd);
1275 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter)
1277 struct kvm_irq_routing_entry kroute = {};
1278 int virq;
1280 if (!kvm_gsi_routing_enabled()) {
1281 return -ENOSYS;
1284 virq = kvm_irqchip_get_virq(s);
1285 if (virq < 0) {
1286 return virq;
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;
1298 kvm_add_routing_entry(s, &kroute);
1300 return virq;
1303 int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint)
1305 struct kvm_irq_routing_entry kroute = {};
1306 int virq;
1308 if (!kvm_gsi_routing_enabled()) {
1309 return -ENOSYS;
1311 if (!kvm_check_extension(s, KVM_CAP_HYPERV_SYNIC)) {
1312 return -ENOSYS;
1314 virq = kvm_irqchip_get_virq(s);
1315 if (virq < 0) {
1316 return virq;
1319 kroute.gsi = virq;
1320 kroute.type = KVM_IRQ_ROUTING_HV_SINT;
1321 kroute.flags = 0;
1322 kroute.u.hv_sint.vcpu = vcpu;
1323 kroute.u.hv_sint.sint = sint;
1325 kvm_add_routing_entry(s, &kroute);
1326 kvm_irqchip_commit_routes(s);
1328 return virq;
1331 #else /* !KVM_CAP_IRQ_ROUTING */
1333 void kvm_init_irq_routing(KVMState *s)
1337 void kvm_irqchip_release_virq(KVMState *s, int virq)
1341 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1343 abort();
1346 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1348 return -ENOSYS;
1351 int kvm_irqchip_add_adapter_route(KVMState *s, AdapterInfo *adapter)
1353 return -ENOSYS;
1356 int kvm_irqchip_add_hv_sint_route(KVMState *s, uint32_t vcpu, uint32_t sint)
1358 return -ENOSYS;
1361 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign)
1363 abort();
1366 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1368 return -ENOSYS;
1370 #endif /* !KVM_CAP_IRQ_ROUTING */
1372 int kvm_irqchip_add_irqfd_notifier_gsi(KVMState *s, EventNotifier *n,
1373 EventNotifier *rn, int virq)
1375 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n),
1376 rn ? event_notifier_get_fd(rn) : -1, virq, true);
1379 int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState *s, EventNotifier *n,
1380 int virq)
1382 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), -1, virq,
1383 false);
1386 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n,
1387 EventNotifier *rn, qemu_irq irq)
1389 gpointer key, gsi;
1390 gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi);
1392 if (!found) {
1393 return -ENXIO;
1395 return kvm_irqchip_add_irqfd_notifier_gsi(s, n, rn, GPOINTER_TO_INT(gsi));
1398 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n,
1399 qemu_irq irq)
1401 gpointer key, gsi;
1402 gboolean found = g_hash_table_lookup_extended(s->gsimap, irq, &key, &gsi);
1404 if (!found) {
1405 return -ENXIO;
1407 return kvm_irqchip_remove_irqfd_notifier_gsi(s, n, GPOINTER_TO_INT(gsi));
1410 void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi)
1412 g_hash_table_insert(s->gsimap, irq, GINT_TO_POINTER(gsi));
1415 static void kvm_irqchip_create(MachineState *machine, KVMState *s)
1417 int ret;
1419 if (kvm_check_extension(s, KVM_CAP_IRQCHIP)) {
1421 } else if (kvm_check_extension(s, KVM_CAP_S390_IRQCHIP)) {
1422 ret = kvm_vm_enable_cap(s, KVM_CAP_S390_IRQCHIP, 0);
1423 if (ret < 0) {
1424 fprintf(stderr, "Enable kernel irqchip failed: %s\n", strerror(-ret));
1425 exit(1);
1427 } else {
1428 return;
1431 /* First probe and see if there's a arch-specific hook to create the
1432 * in-kernel irqchip for us */
1433 ret = kvm_arch_irqchip_create(machine, s);
1434 if (ret == 0) {
1435 if (machine_kernel_irqchip_split(machine)) {
1436 perror("Split IRQ chip mode not supported.");
1437 exit(1);
1438 } else {
1439 ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP);
1442 if (ret < 0) {
1443 fprintf(stderr, "Create kernel irqchip failed: %s\n", strerror(-ret));
1444 exit(1);
1447 kvm_kernel_irqchip = true;
1448 /* If we have an in-kernel IRQ chip then we must have asynchronous
1449 * interrupt delivery (though the reverse is not necessarily true)
1451 kvm_async_interrupts_allowed = true;
1452 kvm_halt_in_kernel_allowed = true;
1454 kvm_init_irq_routing(s);
1456 s->gsimap = g_hash_table_new(g_direct_hash, g_direct_equal);
1459 /* Find number of supported CPUs using the recommended
1460 * procedure from the kernel API documentation to cope with
1461 * older kernels that may be missing capabilities.
1463 static int kvm_recommended_vcpus(KVMState *s)
1465 int ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS);
1466 return (ret) ? ret : 4;
1469 static int kvm_max_vcpus(KVMState *s)
1471 int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS);
1472 return (ret) ? ret : kvm_recommended_vcpus(s);
1475 static int kvm_init(MachineState *ms)
1477 MachineClass *mc = MACHINE_GET_CLASS(ms);
1478 static const char upgrade_note[] =
1479 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1480 "(see http://sourceforge.net/projects/kvm).\n";
1481 struct {
1482 const char *name;
1483 int num;
1484 } num_cpus[] = {
1485 { "SMP", smp_cpus },
1486 { "hotpluggable", max_cpus },
1487 { NULL, }
1488 }, *nc = num_cpus;
1489 int soft_vcpus_limit, hard_vcpus_limit;
1490 KVMState *s;
1491 const KVMCapabilityInfo *missing_cap;
1492 int ret;
1493 int type = 0;
1494 const char *kvm_type;
1496 s = KVM_STATE(ms->accelerator);
1499 * On systems where the kernel can support different base page
1500 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1501 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1502 * page size for the system though.
1504 assert(TARGET_PAGE_SIZE <= getpagesize());
1506 s->sigmask_len = 8;
1508 #ifdef KVM_CAP_SET_GUEST_DEBUG
1509 QTAILQ_INIT(&s->kvm_sw_breakpoints);
1510 #endif
1511 s->vmfd = -1;
1512 s->fd = qemu_open("/dev/kvm", O_RDWR);
1513 if (s->fd == -1) {
1514 fprintf(stderr, "Could not access KVM kernel module: %m\n");
1515 ret = -errno;
1516 goto err;
1519 ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
1520 if (ret < KVM_API_VERSION) {
1521 if (ret >= 0) {
1522 ret = -EINVAL;
1524 fprintf(stderr, "kvm version too old\n");
1525 goto err;
1528 if (ret > KVM_API_VERSION) {
1529 ret = -EINVAL;
1530 fprintf(stderr, "kvm version not supported\n");
1531 goto err;
1534 s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
1536 /* If unspecified, use the default value */
1537 if (!s->nr_slots) {
1538 s->nr_slots = 32;
1541 /* check the vcpu limits */
1542 soft_vcpus_limit = kvm_recommended_vcpus(s);
1543 hard_vcpus_limit = kvm_max_vcpus(s);
1545 while (nc->name) {
1546 if (nc->num > soft_vcpus_limit) {
1547 fprintf(stderr,
1548 "Warning: Number of %s cpus requested (%d) exceeds "
1549 "the recommended cpus supported by KVM (%d)\n",
1550 nc->name, nc->num, soft_vcpus_limit);
1552 if (nc->num > hard_vcpus_limit) {
1553 fprintf(stderr, "Number of %s cpus requested (%d) exceeds "
1554 "the maximum cpus supported by KVM (%d)\n",
1555 nc->name, nc->num, hard_vcpus_limit);
1556 exit(1);
1559 nc++;
1562 kvm_type = qemu_opt_get(qemu_get_machine_opts(), "kvm-type");
1563 if (mc->kvm_type) {
1564 type = mc->kvm_type(kvm_type);
1565 } else if (kvm_type) {
1566 ret = -EINVAL;
1567 fprintf(stderr, "Invalid argument kvm-type=%s\n", kvm_type);
1568 goto err;
1571 do {
1572 ret = kvm_ioctl(s, KVM_CREATE_VM, type);
1573 } while (ret == -EINTR);
1575 if (ret < 0) {
1576 fprintf(stderr, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret,
1577 strerror(-ret));
1579 #ifdef TARGET_S390X
1580 if (ret == -EINVAL) {
1581 fprintf(stderr,
1582 "Host kernel setup problem detected. Please verify:\n");
1583 fprintf(stderr, "- for kernels supporting the switch_amode or"
1584 " user_mode parameters, whether\n");
1585 fprintf(stderr,
1586 " user space is running in primary address space\n");
1587 fprintf(stderr,
1588 "- for kernels supporting the vm.allocate_pgste sysctl, "
1589 "whether it is enabled\n");
1591 #endif
1592 goto err;
1595 s->vmfd = ret;
1596 missing_cap = kvm_check_extension_list(s, kvm_required_capabilites);
1597 if (!missing_cap) {
1598 missing_cap =
1599 kvm_check_extension_list(s, kvm_arch_required_capabilities);
1601 if (missing_cap) {
1602 ret = -EINVAL;
1603 fprintf(stderr, "kvm does not support %s\n%s",
1604 missing_cap->name, upgrade_note);
1605 goto err;
1608 s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
1610 s->broken_set_mem_region = 1;
1611 ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
1612 if (ret > 0) {
1613 s->broken_set_mem_region = 0;
1616 #ifdef KVM_CAP_VCPU_EVENTS
1617 s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS);
1618 #endif
1620 s->robust_singlestep =
1621 kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP);
1623 #ifdef KVM_CAP_DEBUGREGS
1624 s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS);
1625 #endif
1627 #ifdef KVM_CAP_IRQ_ROUTING
1628 kvm_direct_msi_allowed = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0);
1629 #endif
1631 s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3);
1633 s->irq_set_ioctl = KVM_IRQ_LINE;
1634 if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) {
1635 s->irq_set_ioctl = KVM_IRQ_LINE_STATUS;
1638 #ifdef KVM_CAP_READONLY_MEM
1639 kvm_readonly_mem_allowed =
1640 (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0);
1641 #endif
1643 kvm_eventfds_allowed =
1644 (kvm_check_extension(s, KVM_CAP_IOEVENTFD) > 0);
1646 kvm_irqfds_allowed =
1647 (kvm_check_extension(s, KVM_CAP_IRQFD) > 0);
1649 kvm_resamplefds_allowed =
1650 (kvm_check_extension(s, KVM_CAP_IRQFD_RESAMPLE) > 0);
1652 kvm_vm_attributes_allowed =
1653 (kvm_check_extension(s, KVM_CAP_VM_ATTRIBUTES) > 0);
1655 kvm_ioeventfd_any_length_allowed =
1656 (kvm_check_extension(s, KVM_CAP_IOEVENTFD_ANY_LENGTH) > 0);
1658 ret = kvm_arch_init(ms, s);
1659 if (ret < 0) {
1660 goto err;
1663 if (machine_kernel_irqchip_allowed(ms)) {
1664 kvm_irqchip_create(ms, s);
1667 kvm_state = s;
1669 if (kvm_eventfds_allowed) {
1670 s->memory_listener.listener.eventfd_add = kvm_mem_ioeventfd_add;
1671 s->memory_listener.listener.eventfd_del = kvm_mem_ioeventfd_del;
1673 s->memory_listener.listener.coalesced_mmio_add = kvm_coalesce_mmio_region;
1674 s->memory_listener.listener.coalesced_mmio_del = kvm_uncoalesce_mmio_region;
1676 kvm_memory_listener_register(s, &s->memory_listener,
1677 &address_space_memory, 0);
1678 memory_listener_register(&kvm_io_listener,
1679 &address_space_io);
1681 s->many_ioeventfds = kvm_check_many_ioeventfds();
1683 cpu_interrupt_handler = kvm_handle_interrupt;
1685 return 0;
1687 err:
1688 assert(ret < 0);
1689 if (s->vmfd >= 0) {
1690 close(s->vmfd);
1692 if (s->fd != -1) {
1693 close(s->fd);
1695 g_free(s->memory_listener.slots);
1697 return ret;
1700 void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len)
1702 s->sigmask_len = sigmask_len;
1705 static void kvm_handle_io(uint16_t port, MemTxAttrs attrs, void *data, int direction,
1706 int size, uint32_t count)
1708 int i;
1709 uint8_t *ptr = data;
1711 for (i = 0; i < count; i++) {
1712 address_space_rw(&address_space_io, port, attrs,
1713 ptr, size,
1714 direction == KVM_EXIT_IO_OUT);
1715 ptr += size;
1719 static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run)
1721 fprintf(stderr, "KVM internal error. Suberror: %d\n",
1722 run->internal.suberror);
1724 if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) {
1725 int i;
1727 for (i = 0; i < run->internal.ndata; ++i) {
1728 fprintf(stderr, "extra data[%d]: %"PRIx64"\n",
1729 i, (uint64_t)run->internal.data[i]);
1732 if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) {
1733 fprintf(stderr, "emulation failure\n");
1734 if (!kvm_arch_stop_on_emulation_error(cpu)) {
1735 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1736 return EXCP_INTERRUPT;
1739 /* FIXME: Should trigger a qmp message to let management know
1740 * something went wrong.
1742 return -1;
1745 void kvm_flush_coalesced_mmio_buffer(void)
1747 KVMState *s = kvm_state;
1749 if (s->coalesced_flush_in_progress) {
1750 return;
1753 s->coalesced_flush_in_progress = true;
1755 if (s->coalesced_mmio_ring) {
1756 struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring;
1757 while (ring->first != ring->last) {
1758 struct kvm_coalesced_mmio *ent;
1760 ent = &ring->coalesced_mmio[ring->first];
1762 cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
1763 smp_wmb();
1764 ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
1768 s->coalesced_flush_in_progress = false;
1771 static void do_kvm_cpu_synchronize_state(void *arg)
1773 CPUState *cpu = arg;
1775 if (!cpu->kvm_vcpu_dirty) {
1776 kvm_arch_get_registers(cpu);
1777 cpu->kvm_vcpu_dirty = true;
1781 void kvm_cpu_synchronize_state(CPUState *cpu)
1783 if (!cpu->kvm_vcpu_dirty) {
1784 run_on_cpu(cpu, do_kvm_cpu_synchronize_state, cpu);
1788 static void do_kvm_cpu_synchronize_post_reset(void *arg)
1790 CPUState *cpu = arg;
1792 kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE);
1793 cpu->kvm_vcpu_dirty = false;
1796 void kvm_cpu_synchronize_post_reset(CPUState *cpu)
1798 run_on_cpu(cpu, do_kvm_cpu_synchronize_post_reset, cpu);
1801 static void do_kvm_cpu_synchronize_post_init(void *arg)
1803 CPUState *cpu = arg;
1805 kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE);
1806 cpu->kvm_vcpu_dirty = false;
1809 void kvm_cpu_synchronize_post_init(CPUState *cpu)
1811 run_on_cpu(cpu, do_kvm_cpu_synchronize_post_init, cpu);
1814 int kvm_cpu_exec(CPUState *cpu)
1816 struct kvm_run *run = cpu->kvm_run;
1817 int ret, run_ret;
1819 DPRINTF("kvm_cpu_exec()\n");
1821 if (kvm_arch_process_async_events(cpu)) {
1822 cpu->exit_request = 0;
1823 return EXCP_HLT;
1826 qemu_mutex_unlock_iothread();
1828 do {
1829 MemTxAttrs attrs;
1831 if (cpu->kvm_vcpu_dirty) {
1832 kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE);
1833 cpu->kvm_vcpu_dirty = false;
1836 kvm_arch_pre_run(cpu, run);
1837 if (cpu->exit_request) {
1838 DPRINTF("interrupt exit requested\n");
1840 * KVM requires us to reenter the kernel after IO exits to complete
1841 * instruction emulation. This self-signal will ensure that we
1842 * leave ASAP again.
1844 qemu_cpu_kick_self();
1847 run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0);
1849 attrs = kvm_arch_post_run(cpu, run);
1851 if (run_ret < 0) {
1852 if (run_ret == -EINTR || run_ret == -EAGAIN) {
1853 DPRINTF("io window exit\n");
1854 ret = EXCP_INTERRUPT;
1855 break;
1857 fprintf(stderr, "error: kvm run failed %s\n",
1858 strerror(-run_ret));
1859 #ifdef TARGET_PPC
1860 if (run_ret == -EBUSY) {
1861 fprintf(stderr,
1862 "This is probably because your SMT is enabled.\n"
1863 "VCPU can only run on primary threads with all "
1864 "secondary threads offline.\n");
1866 #endif
1867 ret = -1;
1868 break;
1871 trace_kvm_run_exit(cpu->cpu_index, run->exit_reason);
1872 switch (run->exit_reason) {
1873 case KVM_EXIT_IO:
1874 DPRINTF("handle_io\n");
1875 /* Called outside BQL */
1876 kvm_handle_io(run->io.port, attrs,
1877 (uint8_t *)run + run->io.data_offset,
1878 run->io.direction,
1879 run->io.size,
1880 run->io.count);
1881 ret = 0;
1882 break;
1883 case KVM_EXIT_MMIO:
1884 DPRINTF("handle_mmio\n");
1885 /* Called outside BQL */
1886 address_space_rw(&address_space_memory,
1887 run->mmio.phys_addr, attrs,
1888 run->mmio.data,
1889 run->mmio.len,
1890 run->mmio.is_write);
1891 ret = 0;
1892 break;
1893 case KVM_EXIT_IRQ_WINDOW_OPEN:
1894 DPRINTF("irq_window_open\n");
1895 ret = EXCP_INTERRUPT;
1896 break;
1897 case KVM_EXIT_SHUTDOWN:
1898 DPRINTF("shutdown\n");
1899 qemu_system_reset_request();
1900 ret = EXCP_INTERRUPT;
1901 break;
1902 case KVM_EXIT_UNKNOWN:
1903 fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n",
1904 (uint64_t)run->hw.hardware_exit_reason);
1905 ret = -1;
1906 break;
1907 case KVM_EXIT_INTERNAL_ERROR:
1908 ret = kvm_handle_internal_error(cpu, run);
1909 break;
1910 case KVM_EXIT_SYSTEM_EVENT:
1911 switch (run->system_event.type) {
1912 case KVM_SYSTEM_EVENT_SHUTDOWN:
1913 qemu_system_shutdown_request();
1914 ret = EXCP_INTERRUPT;
1915 break;
1916 case KVM_SYSTEM_EVENT_RESET:
1917 qemu_system_reset_request();
1918 ret = EXCP_INTERRUPT;
1919 break;
1920 case KVM_SYSTEM_EVENT_CRASH:
1921 qemu_mutex_lock_iothread();
1922 qemu_system_guest_panicked();
1923 qemu_mutex_unlock_iothread();
1924 ret = 0;
1925 break;
1926 default:
1927 DPRINTF("kvm_arch_handle_exit\n");
1928 ret = kvm_arch_handle_exit(cpu, run);
1929 break;
1931 break;
1932 default:
1933 DPRINTF("kvm_arch_handle_exit\n");
1934 ret = kvm_arch_handle_exit(cpu, run);
1935 break;
1937 } while (ret == 0);
1939 qemu_mutex_lock_iothread();
1941 if (ret < 0) {
1942 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1943 vm_stop(RUN_STATE_INTERNAL_ERROR);
1946 cpu->exit_request = 0;
1947 return ret;
1950 int kvm_ioctl(KVMState *s, int type, ...)
1952 int ret;
1953 void *arg;
1954 va_list ap;
1956 va_start(ap, type);
1957 arg = va_arg(ap, void *);
1958 va_end(ap);
1960 trace_kvm_ioctl(type, arg);
1961 ret = ioctl(s->fd, type, arg);
1962 if (ret == -1) {
1963 ret = -errno;
1965 return ret;
1968 int kvm_vm_ioctl(KVMState *s, int type, ...)
1970 int ret;
1971 void *arg;
1972 va_list ap;
1974 va_start(ap, type);
1975 arg = va_arg(ap, void *);
1976 va_end(ap);
1978 trace_kvm_vm_ioctl(type, arg);
1979 ret = ioctl(s->vmfd, type, arg);
1980 if (ret == -1) {
1981 ret = -errno;
1983 return ret;
1986 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...)
1988 int ret;
1989 void *arg;
1990 va_list ap;
1992 va_start(ap, type);
1993 arg = va_arg(ap, void *);
1994 va_end(ap);
1996 trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg);
1997 ret = ioctl(cpu->kvm_fd, type, arg);
1998 if (ret == -1) {
1999 ret = -errno;
2001 return ret;
2004 int kvm_device_ioctl(int fd, int type, ...)
2006 int ret;
2007 void *arg;
2008 va_list ap;
2010 va_start(ap, type);
2011 arg = va_arg(ap, void *);
2012 va_end(ap);
2014 trace_kvm_device_ioctl(fd, type, arg);
2015 ret = ioctl(fd, type, arg);
2016 if (ret == -1) {
2017 ret = -errno;
2019 return ret;
2022 int kvm_vm_check_attr(KVMState *s, uint32_t group, uint64_t attr)
2024 int ret;
2025 struct kvm_device_attr attribute = {
2026 .group = group,
2027 .attr = attr,
2030 if (!kvm_vm_attributes_allowed) {
2031 return 0;
2034 ret = kvm_vm_ioctl(s, KVM_HAS_DEVICE_ATTR, &attribute);
2035 /* kvm returns 0 on success for HAS_DEVICE_ATTR */
2036 return ret ? 0 : 1;
2039 int kvm_device_check_attr(int dev_fd, uint32_t group, uint64_t attr)
2041 struct kvm_device_attr attribute = {
2042 .group = group,
2043 .attr = attr,
2044 .flags = 0,
2047 return kvm_device_ioctl(dev_fd, KVM_HAS_DEVICE_ATTR, &attribute) ? 0 : 1;
2050 void kvm_device_access(int fd, int group, uint64_t attr,
2051 void *val, bool write)
2053 struct kvm_device_attr kvmattr;
2054 int err;
2056 kvmattr.flags = 0;
2057 kvmattr.group = group;
2058 kvmattr.attr = attr;
2059 kvmattr.addr = (uintptr_t)val;
2061 err = kvm_device_ioctl(fd,
2062 write ? KVM_SET_DEVICE_ATTR : KVM_GET_DEVICE_ATTR,
2063 &kvmattr);
2064 if (err < 0) {
2065 error_report("KVM_%s_DEVICE_ATTR failed: %s",
2066 write ? "SET" : "GET", strerror(-err));
2067 error_printf("Group %d attr 0x%016" PRIx64, group, attr);
2068 abort();
2072 int kvm_has_sync_mmu(void)
2074 return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU);
2077 int kvm_has_vcpu_events(void)
2079 return kvm_state->vcpu_events;
2082 int kvm_has_robust_singlestep(void)
2084 return kvm_state->robust_singlestep;
2087 int kvm_has_debugregs(void)
2089 return kvm_state->debugregs;
2092 int kvm_has_many_ioeventfds(void)
2094 if (!kvm_enabled()) {
2095 return 0;
2097 return kvm_state->many_ioeventfds;
2100 int kvm_has_gsi_routing(void)
2102 #ifdef KVM_CAP_IRQ_ROUTING
2103 return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING);
2104 #else
2105 return false;
2106 #endif
2109 int kvm_has_intx_set_mask(void)
2111 return kvm_state->intx_set_mask;
2114 void kvm_setup_guest_memory(void *start, size_t size)
2116 if (!kvm_has_sync_mmu()) {
2117 int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK);
2119 if (ret) {
2120 perror("qemu_madvise");
2121 fprintf(stderr,
2122 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
2123 exit(1);
2128 #ifdef KVM_CAP_SET_GUEST_DEBUG
2129 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu,
2130 target_ulong pc)
2132 struct kvm_sw_breakpoint *bp;
2134 QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) {
2135 if (bp->pc == pc) {
2136 return bp;
2139 return NULL;
2142 int kvm_sw_breakpoints_active(CPUState *cpu)
2144 return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints);
2147 struct kvm_set_guest_debug_data {
2148 struct kvm_guest_debug dbg;
2149 CPUState *cpu;
2150 int err;
2153 static void kvm_invoke_set_guest_debug(void *data)
2155 struct kvm_set_guest_debug_data *dbg_data = data;
2157 dbg_data->err = kvm_vcpu_ioctl(dbg_data->cpu, KVM_SET_GUEST_DEBUG,
2158 &dbg_data->dbg);
2161 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2163 struct kvm_set_guest_debug_data data;
2165 data.dbg.control = reinject_trap;
2167 if (cpu->singlestep_enabled) {
2168 data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
2170 kvm_arch_update_guest_debug(cpu, &data.dbg);
2171 data.cpu = cpu;
2173 run_on_cpu(cpu, kvm_invoke_set_guest_debug, &data);
2174 return data.err;
2177 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2178 target_ulong len, int type)
2180 struct kvm_sw_breakpoint *bp;
2181 int err;
2183 if (type == GDB_BREAKPOINT_SW) {
2184 bp = kvm_find_sw_breakpoint(cpu, addr);
2185 if (bp) {
2186 bp->use_count++;
2187 return 0;
2190 bp = g_malloc(sizeof(struct kvm_sw_breakpoint));
2191 bp->pc = addr;
2192 bp->use_count = 1;
2193 err = kvm_arch_insert_sw_breakpoint(cpu, bp);
2194 if (err) {
2195 g_free(bp);
2196 return err;
2199 QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
2200 } else {
2201 err = kvm_arch_insert_hw_breakpoint(addr, len, type);
2202 if (err) {
2203 return err;
2207 CPU_FOREACH(cpu) {
2208 err = kvm_update_guest_debug(cpu, 0);
2209 if (err) {
2210 return err;
2213 return 0;
2216 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2217 target_ulong len, int type)
2219 struct kvm_sw_breakpoint *bp;
2220 int err;
2222 if (type == GDB_BREAKPOINT_SW) {
2223 bp = kvm_find_sw_breakpoint(cpu, addr);
2224 if (!bp) {
2225 return -ENOENT;
2228 if (bp->use_count > 1) {
2229 bp->use_count--;
2230 return 0;
2233 err = kvm_arch_remove_sw_breakpoint(cpu, bp);
2234 if (err) {
2235 return err;
2238 QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
2239 g_free(bp);
2240 } else {
2241 err = kvm_arch_remove_hw_breakpoint(addr, len, type);
2242 if (err) {
2243 return err;
2247 CPU_FOREACH(cpu) {
2248 err = kvm_update_guest_debug(cpu, 0);
2249 if (err) {
2250 return err;
2253 return 0;
2256 void kvm_remove_all_breakpoints(CPUState *cpu)
2258 struct kvm_sw_breakpoint *bp, *next;
2259 KVMState *s = cpu->kvm_state;
2260 CPUState *tmpcpu;
2262 QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
2263 if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) {
2264 /* Try harder to find a CPU that currently sees the breakpoint. */
2265 CPU_FOREACH(tmpcpu) {
2266 if (kvm_arch_remove_sw_breakpoint(tmpcpu, bp) == 0) {
2267 break;
2271 QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry);
2272 g_free(bp);
2274 kvm_arch_remove_all_hw_breakpoints();
2276 CPU_FOREACH(cpu) {
2277 kvm_update_guest_debug(cpu, 0);
2281 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2283 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2285 return -EINVAL;
2288 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2289 target_ulong len, int type)
2291 return -EINVAL;
2294 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2295 target_ulong len, int type)
2297 return -EINVAL;
2300 void kvm_remove_all_breakpoints(CPUState *cpu)
2303 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2305 int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset)
2307 KVMState *s = kvm_state;
2308 struct kvm_signal_mask *sigmask;
2309 int r;
2311 if (!sigset) {
2312 return kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, NULL);
2315 sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset));
2317 sigmask->len = s->sigmask_len;
2318 memcpy(sigmask->sigset, sigset, sizeof(*sigset));
2319 r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask);
2320 g_free(sigmask);
2322 return r;
2324 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2326 return kvm_arch_on_sigbus_vcpu(cpu, code, addr);
2329 int kvm_on_sigbus(int code, void *addr)
2331 return kvm_arch_on_sigbus(code, addr);
2334 int kvm_create_device(KVMState *s, uint64_t type, bool test)
2336 int ret;
2337 struct kvm_create_device create_dev;
2339 create_dev.type = type;
2340 create_dev.fd = -1;
2341 create_dev.flags = test ? KVM_CREATE_DEVICE_TEST : 0;
2343 if (!kvm_check_extension(s, KVM_CAP_DEVICE_CTRL)) {
2344 return -ENOTSUP;
2347 ret = kvm_vm_ioctl(s, KVM_CREATE_DEVICE, &create_dev);
2348 if (ret) {
2349 return ret;
2352 return test ? 0 : create_dev.fd;
2355 int kvm_set_one_reg(CPUState *cs, uint64_t id, void *source)
2357 struct kvm_one_reg reg;
2358 int r;
2360 reg.id = id;
2361 reg.addr = (uintptr_t) source;
2362 r = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
2363 if (r) {
2364 trace_kvm_failed_reg_set(id, strerror(r));
2366 return r;
2369 int kvm_get_one_reg(CPUState *cs, uint64_t id, void *target)
2371 struct kvm_one_reg reg;
2372 int r;
2374 reg.id = id;
2375 reg.addr = (uintptr_t) target;
2376 r = kvm_vcpu_ioctl(cs, KVM_GET_ONE_REG, &reg);
2377 if (r) {
2378 trace_kvm_failed_reg_get(id, strerror(r));
2380 return r;
2383 static void kvm_accel_class_init(ObjectClass *oc, void *data)
2385 AccelClass *ac = ACCEL_CLASS(oc);
2386 ac->name = "KVM";
2387 ac->init_machine = kvm_init;
2388 ac->allowed = &kvm_allowed;
2391 static const TypeInfo kvm_accel_type = {
2392 .name = TYPE_KVM_ACCEL,
2393 .parent = TYPE_ACCEL,
2394 .class_init = kvm_accel_class_init,
2395 .instance_size = sizeof(KVMState),
2398 static void kvm_type_init(void)
2400 type_register_static(&kvm_accel_type);
2403 type_init(kvm_type_init);