4 * Copyright IBM, Corp. 2008
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>
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 "sysemu/sysemu.h"
29 #include "hw/pci/msi.h"
30 #include "exec/gdbstub.h"
31 #include "sysemu/kvm.h"
32 #include "qemu/bswap.h"
33 #include "exec/memory.h"
34 #include "exec/ram_addr.h"
35 #include "exec/address-spaces.h"
36 #include "qemu/event_notifier.h"
39 #include "hw/boards.h"
41 /* This check must be after config-host.h is included */
43 #include <sys/eventfd.h>
46 #ifdef CONFIG_VALGRIND_H
47 #include <valgrind/memcheck.h>
50 /* KVM uses PAGE_SIZE in its definition of COALESCED_MMIO_MAX */
51 #define PAGE_SIZE TARGET_PAGE_SIZE
56 #define DPRINTF(fmt, ...) \
57 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
59 #define DPRINTF(fmt, ...) \
63 #define KVM_MSI_HASHTAB_SIZE 256
65 typedef struct KVMSlot
68 ram_addr_t memory_size;
74 typedef struct kvm_dirty_log KVMDirtyLog;
83 struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
84 bool coalesced_flush_in_progress;
85 int broken_set_mem_region;
88 int robust_singlestep;
90 #ifdef KVM_CAP_SET_GUEST_DEBUG
91 struct kvm_sw_breakpoint_head kvm_sw_breakpoints;
97 /* The man page (and posix) say ioctl numbers are signed int, but
98 * they're not. Linux, glibc and *BSD all treat ioctl numbers as
99 * unsigned, and treating them as signed here can break things */
100 unsigned irq_set_ioctl;
101 #ifdef KVM_CAP_IRQ_ROUTING
102 struct kvm_irq_routing *irq_routes;
103 int nr_allocated_irq_routes;
104 uint32_t *used_gsi_bitmap;
105 unsigned int gsi_count;
106 QTAILQ_HEAD(msi_hashtab, KVMMSIRoute) msi_hashtab[KVM_MSI_HASHTAB_SIZE];
112 bool kvm_kernel_irqchip;
113 bool kvm_async_interrupts_allowed;
114 bool kvm_halt_in_kernel_allowed;
115 bool kvm_irqfds_allowed;
116 bool kvm_msi_via_irqfd_allowed;
117 bool kvm_gsi_routing_allowed;
118 bool kvm_gsi_direct_mapping;
120 bool kvm_readonly_mem_allowed;
122 static const KVMCapabilityInfo kvm_required_capabilites[] = {
123 KVM_CAP_INFO(USER_MEMORY),
124 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS),
128 static KVMSlot *kvm_alloc_slot(KVMState *s)
132 for (i = 0; i < s->nr_slots; i++) {
133 if (s->slots[i].memory_size == 0) {
138 fprintf(stderr, "%s: no free slot available\n", __func__);
142 static KVMSlot *kvm_lookup_matching_slot(KVMState *s,
148 for (i = 0; i < s->nr_slots; i++) {
149 KVMSlot *mem = &s->slots[i];
151 if (start_addr == mem->start_addr &&
152 end_addr == mem->start_addr + mem->memory_size) {
161 * Find overlapping slot with lowest start address
163 static KVMSlot *kvm_lookup_overlapping_slot(KVMState *s,
167 KVMSlot *found = NULL;
170 for (i = 0; i < s->nr_slots; i++) {
171 KVMSlot *mem = &s->slots[i];
173 if (mem->memory_size == 0 ||
174 (found && found->start_addr < mem->start_addr)) {
178 if (end_addr > mem->start_addr &&
179 start_addr < mem->start_addr + mem->memory_size) {
187 int kvm_physical_memory_addr_from_host(KVMState *s, void *ram,
192 for (i = 0; i < s->nr_slots; i++) {
193 KVMSlot *mem = &s->slots[i];
195 if (ram >= mem->ram && ram < mem->ram + mem->memory_size) {
196 *phys_addr = mem->start_addr + (ram - mem->ram);
204 static int kvm_set_user_memory_region(KVMState *s, KVMSlot *slot)
206 struct kvm_userspace_memory_region mem;
208 mem.slot = slot->slot;
209 mem.guest_phys_addr = slot->start_addr;
210 mem.userspace_addr = (unsigned long)slot->ram;
211 mem.flags = slot->flags;
212 if (s->migration_log) {
213 mem.flags |= KVM_MEM_LOG_DIRTY_PAGES;
216 if (slot->memory_size && mem.flags & KVM_MEM_READONLY) {
217 /* Set the slot size to 0 before setting the slot to the desired
218 * value. This is needed based on KVM commit 75d61fbc. */
220 kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
222 mem.memory_size = slot->memory_size;
223 return kvm_vm_ioctl(s, KVM_SET_USER_MEMORY_REGION, &mem);
226 static void kvm_reset_vcpu(void *opaque)
228 CPUState *cpu = opaque;
230 kvm_arch_reset_vcpu(cpu);
233 int kvm_init_vcpu(CPUState *cpu)
235 KVMState *s = kvm_state;
239 DPRINTF("kvm_init_vcpu\n");
241 ret = kvm_vm_ioctl(s, KVM_CREATE_VCPU, (void *)kvm_arch_vcpu_id(cpu));
243 DPRINTF("kvm_create_vcpu failed\n");
249 cpu->kvm_vcpu_dirty = true;
251 mmap_size = kvm_ioctl(s, KVM_GET_VCPU_MMAP_SIZE, 0);
254 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
258 cpu->kvm_run = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
260 if (cpu->kvm_run == MAP_FAILED) {
262 DPRINTF("mmap'ing vcpu state failed\n");
266 if (s->coalesced_mmio && !s->coalesced_mmio_ring) {
267 s->coalesced_mmio_ring =
268 (void *)cpu->kvm_run + s->coalesced_mmio * PAGE_SIZE;
271 ret = kvm_arch_init_vcpu(cpu);
273 qemu_register_reset(kvm_reset_vcpu, cpu);
274 kvm_arch_reset_vcpu(cpu);
281 * dirty pages logging control
284 static int kvm_mem_flags(KVMState *s, bool log_dirty, bool readonly)
287 flags = log_dirty ? KVM_MEM_LOG_DIRTY_PAGES : 0;
288 if (readonly && kvm_readonly_mem_allowed) {
289 flags |= KVM_MEM_READONLY;
294 static int kvm_slot_dirty_pages_log_change(KVMSlot *mem, bool log_dirty)
296 KVMState *s = kvm_state;
297 int flags, mask = KVM_MEM_LOG_DIRTY_PAGES;
300 old_flags = mem->flags;
302 flags = (mem->flags & ~mask) | kvm_mem_flags(s, log_dirty, false);
305 /* If nothing changed effectively, no need to issue ioctl */
306 if (s->migration_log) {
307 flags |= KVM_MEM_LOG_DIRTY_PAGES;
310 if (flags == old_flags) {
314 return kvm_set_user_memory_region(s, mem);
317 static int kvm_dirty_pages_log_change(hwaddr phys_addr,
318 ram_addr_t size, bool log_dirty)
320 KVMState *s = kvm_state;
321 KVMSlot *mem = kvm_lookup_matching_slot(s, phys_addr, phys_addr + size);
324 fprintf(stderr, "BUG: %s: invalid parameters " TARGET_FMT_plx "-"
325 TARGET_FMT_plx "\n", __func__, phys_addr,
326 (hwaddr)(phys_addr + size - 1));
329 return kvm_slot_dirty_pages_log_change(mem, log_dirty);
332 static void kvm_log_start(MemoryListener *listener,
333 MemoryRegionSection *section)
337 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
338 int128_get64(section->size), true);
344 static void kvm_log_stop(MemoryListener *listener,
345 MemoryRegionSection *section)
349 r = kvm_dirty_pages_log_change(section->offset_within_address_space,
350 int128_get64(section->size), false);
356 static int kvm_set_migration_log(int enable)
358 KVMState *s = kvm_state;
362 s->migration_log = enable;
364 for (i = 0; i < s->nr_slots; i++) {
367 if (!mem->memory_size) {
370 if (!!(mem->flags & KVM_MEM_LOG_DIRTY_PAGES) == enable) {
373 err = kvm_set_user_memory_region(s, mem);
381 /* get kvm's dirty pages bitmap and update qemu's */
382 static int kvm_get_dirty_pages_log_range(MemoryRegionSection *section,
383 unsigned long *bitmap)
385 ram_addr_t start = section->offset_within_region + section->mr->ram_addr;
386 ram_addr_t pages = int128_get64(section->size) / getpagesize();
388 cpu_physical_memory_set_dirty_lebitmap(bitmap, start, pages);
392 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
395 * kvm_physical_sync_dirty_bitmap - Grab dirty bitmap from kernel space
396 * This function updates qemu's dirty bitmap using
397 * memory_region_set_dirty(). This means all bits are set
400 * @start_add: start of logged region.
401 * @end_addr: end of logged region.
403 static int kvm_physical_sync_dirty_bitmap(MemoryRegionSection *section)
405 KVMState *s = kvm_state;
406 unsigned long size, allocated_size = 0;
410 hwaddr start_addr = section->offset_within_address_space;
411 hwaddr end_addr = start_addr + int128_get64(section->size);
413 d.dirty_bitmap = NULL;
414 while (start_addr < end_addr) {
415 mem = kvm_lookup_overlapping_slot(s, start_addr, end_addr);
420 /* XXX bad kernel interface alert
421 * For dirty bitmap, kernel allocates array of size aligned to
422 * bits-per-long. But for case when the kernel is 64bits and
423 * the userspace is 32bits, userspace can't align to the same
424 * bits-per-long, since sizeof(long) is different between kernel
425 * and user space. This way, userspace will provide buffer which
426 * may be 4 bytes less than the kernel will use, resulting in
427 * userspace memory corruption (which is not detectable by valgrind
428 * too, in most cases).
429 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
430 * a hope that sizeof(long) wont become >8 any time soon.
432 size = ALIGN(((mem->memory_size) >> TARGET_PAGE_BITS),
433 /*HOST_LONG_BITS*/ 64) / 8;
434 if (!d.dirty_bitmap) {
435 d.dirty_bitmap = g_malloc(size);
436 } else if (size > allocated_size) {
437 d.dirty_bitmap = g_realloc(d.dirty_bitmap, size);
439 allocated_size = size;
440 memset(d.dirty_bitmap, 0, allocated_size);
444 if (kvm_vm_ioctl(s, KVM_GET_DIRTY_LOG, &d) == -1) {
445 DPRINTF("ioctl failed %d\n", errno);
450 kvm_get_dirty_pages_log_range(section, d.dirty_bitmap);
451 start_addr = mem->start_addr + mem->memory_size;
453 g_free(d.dirty_bitmap);
458 static void kvm_coalesce_mmio_region(MemoryListener *listener,
459 MemoryRegionSection *secion,
460 hwaddr start, hwaddr size)
462 KVMState *s = kvm_state;
464 if (s->coalesced_mmio) {
465 struct kvm_coalesced_mmio_zone zone;
471 (void)kvm_vm_ioctl(s, KVM_REGISTER_COALESCED_MMIO, &zone);
475 static void kvm_uncoalesce_mmio_region(MemoryListener *listener,
476 MemoryRegionSection *secion,
477 hwaddr start, hwaddr size)
479 KVMState *s = kvm_state;
481 if (s->coalesced_mmio) {
482 struct kvm_coalesced_mmio_zone zone;
488 (void)kvm_vm_ioctl(s, KVM_UNREGISTER_COALESCED_MMIO, &zone);
492 int kvm_check_extension(KVMState *s, unsigned int extension)
496 ret = kvm_ioctl(s, KVM_CHECK_EXTENSION, extension);
504 static int kvm_set_ioeventfd_mmio(int fd, hwaddr addr, uint32_t val,
505 bool assign, uint32_t size, bool datamatch)
508 struct kvm_ioeventfd iofd;
510 iofd.datamatch = datamatch ? val : 0;
516 if (!kvm_enabled()) {
521 iofd.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
524 iofd.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
527 ret = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &iofd);
536 static int kvm_set_ioeventfd_pio(int fd, uint16_t addr, uint16_t val,
537 bool assign, uint32_t size, bool datamatch)
539 struct kvm_ioeventfd kick = {
540 .datamatch = datamatch ? val : 0,
542 .flags = KVM_IOEVENTFD_FLAG_PIO,
547 if (!kvm_enabled()) {
551 kick.flags |= KVM_IOEVENTFD_FLAG_DATAMATCH;
554 kick.flags |= KVM_IOEVENTFD_FLAG_DEASSIGN;
556 r = kvm_vm_ioctl(kvm_state, KVM_IOEVENTFD, &kick);
564 static int kvm_check_many_ioeventfds(void)
566 /* Userspace can use ioeventfd for io notification. This requires a host
567 * that supports eventfd(2) and an I/O thread; since eventfd does not
568 * support SIGIO it cannot interrupt the vcpu.
570 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
571 * can avoid creating too many ioeventfds.
573 #if defined(CONFIG_EVENTFD)
576 for (i = 0; i < ARRAY_SIZE(ioeventfds); i++) {
577 ioeventfds[i] = eventfd(0, EFD_CLOEXEC);
578 if (ioeventfds[i] < 0) {
581 ret = kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, true, 2, true);
583 close(ioeventfds[i]);
588 /* Decide whether many devices are supported or not */
589 ret = i == ARRAY_SIZE(ioeventfds);
592 kvm_set_ioeventfd_pio(ioeventfds[i], 0, i, false, 2, true);
593 close(ioeventfds[i]);
601 static const KVMCapabilityInfo *
602 kvm_check_extension_list(KVMState *s, const KVMCapabilityInfo *list)
605 if (!kvm_check_extension(s, list->value)) {
613 static void kvm_set_phys_mem(MemoryRegionSection *section, bool add)
615 KVMState *s = kvm_state;
618 MemoryRegion *mr = section->mr;
619 bool log_dirty = memory_region_is_logging(mr);
620 bool writeable = !mr->readonly && !mr->rom_device;
621 bool readonly_flag = mr->readonly || memory_region_is_romd(mr);
622 hwaddr start_addr = section->offset_within_address_space;
623 ram_addr_t size = int128_get64(section->size);
627 /* kvm works in page size chunks, but the function may be called
628 with sub-page size and unaligned start address. */
629 delta = TARGET_PAGE_ALIGN(size) - size;
635 size &= TARGET_PAGE_MASK;
636 if (!size || (start_addr & ~TARGET_PAGE_MASK)) {
640 if (!memory_region_is_ram(mr)) {
641 if (writeable || !kvm_readonly_mem_allowed) {
643 } else if (!mr->romd_mode) {
644 /* If the memory device is not in romd_mode, then we actually want
645 * to remove the kvm memory slot so all accesses will trap. */
650 ram = memory_region_get_ram_ptr(mr) + section->offset_within_region + delta;
653 mem = kvm_lookup_overlapping_slot(s, start_addr, start_addr + size);
658 if (add && start_addr >= mem->start_addr &&
659 (start_addr + size <= mem->start_addr + mem->memory_size) &&
660 (ram - start_addr == mem->ram - mem->start_addr)) {
661 /* The new slot fits into the existing one and comes with
662 * identical parameters - update flags and done. */
663 kvm_slot_dirty_pages_log_change(mem, log_dirty);
669 if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
670 kvm_physical_sync_dirty_bitmap(section);
673 /* unregister the overlapping slot */
674 mem->memory_size = 0;
675 err = kvm_set_user_memory_region(s, mem);
677 fprintf(stderr, "%s: error unregistering overlapping slot: %s\n",
678 __func__, strerror(-err));
682 /* Workaround for older KVM versions: we can't join slots, even not by
683 * unregistering the previous ones and then registering the larger
684 * slot. We have to maintain the existing fragmentation. Sigh.
686 * This workaround assumes that the new slot starts at the same
687 * address as the first existing one. If not or if some overlapping
688 * slot comes around later, we will fail (not seen in practice so far)
689 * - and actually require a recent KVM version. */
690 if (s->broken_set_mem_region &&
691 old.start_addr == start_addr && old.memory_size < size && add) {
692 mem = kvm_alloc_slot(s);
693 mem->memory_size = old.memory_size;
694 mem->start_addr = old.start_addr;
696 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
698 err = kvm_set_user_memory_region(s, mem);
700 fprintf(stderr, "%s: error updating slot: %s\n", __func__,
705 start_addr += old.memory_size;
706 ram += old.memory_size;
707 size -= old.memory_size;
711 /* register prefix slot */
712 if (old.start_addr < start_addr) {
713 mem = kvm_alloc_slot(s);
714 mem->memory_size = start_addr - old.start_addr;
715 mem->start_addr = old.start_addr;
717 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
719 err = kvm_set_user_memory_region(s, mem);
721 fprintf(stderr, "%s: error registering prefix slot: %s\n",
722 __func__, strerror(-err));
724 fprintf(stderr, "%s: This is probably because your kernel's " \
725 "PAGE_SIZE is too big. Please try to use 4k " \
726 "PAGE_SIZE!\n", __func__);
732 /* register suffix slot */
733 if (old.start_addr + old.memory_size > start_addr + size) {
734 ram_addr_t size_delta;
736 mem = kvm_alloc_slot(s);
737 mem->start_addr = start_addr + size;
738 size_delta = mem->start_addr - old.start_addr;
739 mem->memory_size = old.memory_size - size_delta;
740 mem->ram = old.ram + size_delta;
741 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
743 err = kvm_set_user_memory_region(s, mem);
745 fprintf(stderr, "%s: error registering suffix slot: %s\n",
746 __func__, strerror(-err));
752 /* in case the KVM bug workaround already "consumed" the new slot */
759 mem = kvm_alloc_slot(s);
760 mem->memory_size = size;
761 mem->start_addr = start_addr;
763 mem->flags = kvm_mem_flags(s, log_dirty, readonly_flag);
765 err = kvm_set_user_memory_region(s, mem);
767 fprintf(stderr, "%s: error registering slot: %s\n", __func__,
773 static void kvm_region_add(MemoryListener *listener,
774 MemoryRegionSection *section)
776 memory_region_ref(section->mr);
777 kvm_set_phys_mem(section, true);
780 static void kvm_region_del(MemoryListener *listener,
781 MemoryRegionSection *section)
783 kvm_set_phys_mem(section, false);
784 memory_region_unref(section->mr);
787 static void kvm_log_sync(MemoryListener *listener,
788 MemoryRegionSection *section)
792 r = kvm_physical_sync_dirty_bitmap(section);
798 static void kvm_log_global_start(struct MemoryListener *listener)
802 r = kvm_set_migration_log(1);
806 static void kvm_log_global_stop(struct MemoryListener *listener)
810 r = kvm_set_migration_log(0);
814 static void kvm_mem_ioeventfd_add(MemoryListener *listener,
815 MemoryRegionSection *section,
816 bool match_data, uint64_t data,
819 int fd = event_notifier_get_fd(e);
822 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
823 data, true, int128_get64(section->size),
826 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
827 __func__, strerror(-r));
832 static void kvm_mem_ioeventfd_del(MemoryListener *listener,
833 MemoryRegionSection *section,
834 bool match_data, uint64_t data,
837 int fd = event_notifier_get_fd(e);
840 r = kvm_set_ioeventfd_mmio(fd, section->offset_within_address_space,
841 data, false, int128_get64(section->size),
848 static void kvm_io_ioeventfd_add(MemoryListener *listener,
849 MemoryRegionSection *section,
850 bool match_data, uint64_t data,
853 int fd = event_notifier_get_fd(e);
856 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
857 data, true, int128_get64(section->size),
860 fprintf(stderr, "%s: error adding ioeventfd: %s\n",
861 __func__, strerror(-r));
866 static void kvm_io_ioeventfd_del(MemoryListener *listener,
867 MemoryRegionSection *section,
868 bool match_data, uint64_t data,
872 int fd = event_notifier_get_fd(e);
875 r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
876 data, false, int128_get64(section->size),
883 static MemoryListener kvm_memory_listener = {
884 .region_add = kvm_region_add,
885 .region_del = kvm_region_del,
886 .log_start = kvm_log_start,
887 .log_stop = kvm_log_stop,
888 .log_sync = kvm_log_sync,
889 .log_global_start = kvm_log_global_start,
890 .log_global_stop = kvm_log_global_stop,
891 .eventfd_add = kvm_mem_ioeventfd_add,
892 .eventfd_del = kvm_mem_ioeventfd_del,
893 .coalesced_mmio_add = kvm_coalesce_mmio_region,
894 .coalesced_mmio_del = kvm_uncoalesce_mmio_region,
898 static MemoryListener kvm_io_listener = {
899 .eventfd_add = kvm_io_ioeventfd_add,
900 .eventfd_del = kvm_io_ioeventfd_del,
904 static void kvm_handle_interrupt(CPUState *cpu, int mask)
906 cpu->interrupt_request |= mask;
908 if (!qemu_cpu_is_self(cpu)) {
913 int kvm_set_irq(KVMState *s, int irq, int level)
915 struct kvm_irq_level event;
918 assert(kvm_async_interrupts_enabled());
922 ret = kvm_vm_ioctl(s, s->irq_set_ioctl, &event);
924 perror("kvm_set_irq");
928 return (s->irq_set_ioctl == KVM_IRQ_LINE) ? 1 : event.status;
931 #ifdef KVM_CAP_IRQ_ROUTING
932 typedef struct KVMMSIRoute {
933 struct kvm_irq_routing_entry kroute;
934 QTAILQ_ENTRY(KVMMSIRoute) entry;
937 static void set_gsi(KVMState *s, unsigned int gsi)
939 s->used_gsi_bitmap[gsi / 32] |= 1U << (gsi % 32);
942 static void clear_gsi(KVMState *s, unsigned int gsi)
944 s->used_gsi_bitmap[gsi / 32] &= ~(1U << (gsi % 32));
947 void kvm_init_irq_routing(KVMState *s)
951 gsi_count = kvm_check_extension(s, KVM_CAP_IRQ_ROUTING);
953 unsigned int gsi_bits, i;
955 /* Round up so we can search ints using ffs */
956 gsi_bits = ALIGN(gsi_count, 32);
957 s->used_gsi_bitmap = g_malloc0(gsi_bits / 8);
958 s->gsi_count = gsi_count;
960 /* Mark any over-allocated bits as already in use */
961 for (i = gsi_count; i < gsi_bits; i++) {
966 s->irq_routes = g_malloc0(sizeof(*s->irq_routes));
967 s->nr_allocated_irq_routes = 0;
969 if (!s->direct_msi) {
970 for (i = 0; i < KVM_MSI_HASHTAB_SIZE; i++) {
971 QTAILQ_INIT(&s->msi_hashtab[i]);
975 kvm_arch_init_irq_routing(s);
978 void kvm_irqchip_commit_routes(KVMState *s)
982 s->irq_routes->flags = 0;
983 ret = kvm_vm_ioctl(s, KVM_SET_GSI_ROUTING, s->irq_routes);
987 static void kvm_add_routing_entry(KVMState *s,
988 struct kvm_irq_routing_entry *entry)
990 struct kvm_irq_routing_entry *new;
993 if (s->irq_routes->nr == s->nr_allocated_irq_routes) {
994 n = s->nr_allocated_irq_routes * 2;
998 size = sizeof(struct kvm_irq_routing);
999 size += n * sizeof(*new);
1000 s->irq_routes = g_realloc(s->irq_routes, size);
1001 s->nr_allocated_irq_routes = n;
1003 n = s->irq_routes->nr++;
1004 new = &s->irq_routes->entries[n];
1008 set_gsi(s, entry->gsi);
1011 static int kvm_update_routing_entry(KVMState *s,
1012 struct kvm_irq_routing_entry *new_entry)
1014 struct kvm_irq_routing_entry *entry;
1017 for (n = 0; n < s->irq_routes->nr; n++) {
1018 entry = &s->irq_routes->entries[n];
1019 if (entry->gsi != new_entry->gsi) {
1023 if(!memcmp(entry, new_entry, sizeof *entry)) {
1027 *entry = *new_entry;
1029 kvm_irqchip_commit_routes(s);
1037 void kvm_irqchip_add_irq_route(KVMState *s, int irq, int irqchip, int pin)
1039 struct kvm_irq_routing_entry e = {};
1041 assert(pin < s->gsi_count);
1044 e.type = KVM_IRQ_ROUTING_IRQCHIP;
1046 e.u.irqchip.irqchip = irqchip;
1047 e.u.irqchip.pin = pin;
1048 kvm_add_routing_entry(s, &e);
1051 void kvm_irqchip_release_virq(KVMState *s, int virq)
1053 struct kvm_irq_routing_entry *e;
1056 if (kvm_gsi_direct_mapping()) {
1060 for (i = 0; i < s->irq_routes->nr; i++) {
1061 e = &s->irq_routes->entries[i];
1062 if (e->gsi == virq) {
1063 s->irq_routes->nr--;
1064 *e = s->irq_routes->entries[s->irq_routes->nr];
1070 static unsigned int kvm_hash_msi(uint32_t data)
1072 /* This is optimized for IA32 MSI layout. However, no other arch shall
1073 * repeat the mistake of not providing a direct MSI injection API. */
1077 static void kvm_flush_dynamic_msi_routes(KVMState *s)
1079 KVMMSIRoute *route, *next;
1082 for (hash = 0; hash < KVM_MSI_HASHTAB_SIZE; hash++) {
1083 QTAILQ_FOREACH_SAFE(route, &s->msi_hashtab[hash], entry, next) {
1084 kvm_irqchip_release_virq(s, route->kroute.gsi);
1085 QTAILQ_REMOVE(&s->msi_hashtab[hash], route, entry);
1091 static int kvm_irqchip_get_virq(KVMState *s)
1093 uint32_t *word = s->used_gsi_bitmap;
1094 int max_words = ALIGN(s->gsi_count, 32) / 32;
1099 /* Return the lowest unused GSI in the bitmap */
1100 for (i = 0; i < max_words; i++) {
1101 bit = ffs(~word[i]);
1106 return bit - 1 + i * 32;
1108 if (!s->direct_msi && retry) {
1110 kvm_flush_dynamic_msi_routes(s);
1117 static KVMMSIRoute *kvm_lookup_msi_route(KVMState *s, MSIMessage msg)
1119 unsigned int hash = kvm_hash_msi(msg.data);
1122 QTAILQ_FOREACH(route, &s->msi_hashtab[hash], entry) {
1123 if (route->kroute.u.msi.address_lo == (uint32_t)msg.address &&
1124 route->kroute.u.msi.address_hi == (msg.address >> 32) &&
1125 route->kroute.u.msi.data == le32_to_cpu(msg.data)) {
1132 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1137 if (s->direct_msi) {
1138 msi.address_lo = (uint32_t)msg.address;
1139 msi.address_hi = msg.address >> 32;
1140 msi.data = le32_to_cpu(msg.data);
1142 memset(msi.pad, 0, sizeof(msi.pad));
1144 return kvm_vm_ioctl(s, KVM_SIGNAL_MSI, &msi);
1147 route = kvm_lookup_msi_route(s, msg);
1151 virq = kvm_irqchip_get_virq(s);
1156 route = g_malloc0(sizeof(KVMMSIRoute));
1157 route->kroute.gsi = virq;
1158 route->kroute.type = KVM_IRQ_ROUTING_MSI;
1159 route->kroute.flags = 0;
1160 route->kroute.u.msi.address_lo = (uint32_t)msg.address;
1161 route->kroute.u.msi.address_hi = msg.address >> 32;
1162 route->kroute.u.msi.data = le32_to_cpu(msg.data);
1164 kvm_add_routing_entry(s, &route->kroute);
1165 kvm_irqchip_commit_routes(s);
1167 QTAILQ_INSERT_TAIL(&s->msi_hashtab[kvm_hash_msi(msg.data)], route,
1171 assert(route->kroute.type == KVM_IRQ_ROUTING_MSI);
1173 return kvm_set_irq(s, route->kroute.gsi, 1);
1176 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1178 struct kvm_irq_routing_entry kroute = {};
1181 if (kvm_gsi_direct_mapping()) {
1182 return msg.data & 0xffff;
1185 if (!kvm_gsi_routing_enabled()) {
1189 virq = kvm_irqchip_get_virq(s);
1195 kroute.type = KVM_IRQ_ROUTING_MSI;
1197 kroute.u.msi.address_lo = (uint32_t)msg.address;
1198 kroute.u.msi.address_hi = msg.address >> 32;
1199 kroute.u.msi.data = le32_to_cpu(msg.data);
1201 kvm_add_routing_entry(s, &kroute);
1202 kvm_irqchip_commit_routes(s);
1207 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1209 struct kvm_irq_routing_entry kroute = {};
1211 if (kvm_gsi_direct_mapping()) {
1215 if (!kvm_irqchip_in_kernel()) {
1220 kroute.type = KVM_IRQ_ROUTING_MSI;
1222 kroute.u.msi.address_lo = (uint32_t)msg.address;
1223 kroute.u.msi.address_hi = msg.address >> 32;
1224 kroute.u.msi.data = le32_to_cpu(msg.data);
1226 return kvm_update_routing_entry(s, &kroute);
1229 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int rfd, int virq,
1232 struct kvm_irqfd irqfd = {
1235 .flags = assign ? 0 : KVM_IRQFD_FLAG_DEASSIGN,
1239 irqfd.flags |= KVM_IRQFD_FLAG_RESAMPLE;
1240 irqfd.resamplefd = rfd;
1243 if (!kvm_irqfds_enabled()) {
1247 return kvm_vm_ioctl(s, KVM_IRQFD, &irqfd);
1250 #else /* !KVM_CAP_IRQ_ROUTING */
1252 void kvm_init_irq_routing(KVMState *s)
1256 void kvm_irqchip_release_virq(KVMState *s, int virq)
1260 int kvm_irqchip_send_msi(KVMState *s, MSIMessage msg)
1265 int kvm_irqchip_add_msi_route(KVMState *s, MSIMessage msg)
1270 static int kvm_irqchip_assign_irqfd(KVMState *s, int fd, int virq, bool assign)
1275 int kvm_irqchip_update_msi_route(KVMState *s, int virq, MSIMessage msg)
1279 #endif /* !KVM_CAP_IRQ_ROUTING */
1281 int kvm_irqchip_add_irqfd_notifier(KVMState *s, EventNotifier *n,
1282 EventNotifier *rn, int virq)
1284 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n),
1285 rn ? event_notifier_get_fd(rn) : -1, virq, true);
1288 int kvm_irqchip_remove_irqfd_notifier(KVMState *s, EventNotifier *n, int virq)
1290 return kvm_irqchip_assign_irqfd(s, event_notifier_get_fd(n), -1, virq,
1294 static int kvm_irqchip_create(KVMState *s)
1298 if (!qemu_opt_get_bool(qemu_get_machine_opts(), "kernel_irqchip", true) ||
1299 !kvm_check_extension(s, KVM_CAP_IRQCHIP)) {
1303 /* First probe and see if there's a arch-specific hook to create the
1304 * in-kernel irqchip for us */
1305 ret = kvm_arch_irqchip_create(s);
1308 } else if (ret == 0) {
1309 ret = kvm_vm_ioctl(s, KVM_CREATE_IRQCHIP);
1311 fprintf(stderr, "Create kernel irqchip failed\n");
1316 kvm_kernel_irqchip = true;
1317 /* If we have an in-kernel IRQ chip then we must have asynchronous
1318 * interrupt delivery (though the reverse is not necessarily true)
1320 kvm_async_interrupts_allowed = true;
1321 kvm_halt_in_kernel_allowed = true;
1323 kvm_init_irq_routing(s);
1328 /* Find number of supported CPUs using the recommended
1329 * procedure from the kernel API documentation to cope with
1330 * older kernels that may be missing capabilities.
1332 static int kvm_recommended_vcpus(KVMState *s)
1334 int ret = kvm_check_extension(s, KVM_CAP_NR_VCPUS);
1335 return (ret) ? ret : 4;
1338 static int kvm_max_vcpus(KVMState *s)
1340 int ret = kvm_check_extension(s, KVM_CAP_MAX_VCPUS);
1341 return (ret) ? ret : kvm_recommended_vcpus(s);
1344 int kvm_init(QEMUMachine *machine)
1346 static const char upgrade_note[] =
1347 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1348 "(see http://sourceforge.net/projects/kvm).\n";
1353 { "SMP", smp_cpus },
1354 { "hotpluggable", max_cpus },
1357 int soft_vcpus_limit, hard_vcpus_limit;
1359 const KVMCapabilityInfo *missing_cap;
1362 const char *kvm_type;
1364 s = g_malloc0(sizeof(KVMState));
1367 * On systems where the kernel can support different base page
1368 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1369 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1370 * page size for the system though.
1372 assert(TARGET_PAGE_SIZE <= getpagesize());
1375 #ifdef KVM_CAP_SET_GUEST_DEBUG
1376 QTAILQ_INIT(&s->kvm_sw_breakpoints);
1379 s->fd = qemu_open("/dev/kvm", O_RDWR);
1381 fprintf(stderr, "Could not access KVM kernel module: %m\n");
1386 ret = kvm_ioctl(s, KVM_GET_API_VERSION, 0);
1387 if (ret < KVM_API_VERSION) {
1391 fprintf(stderr, "kvm version too old\n");
1395 if (ret > KVM_API_VERSION) {
1397 fprintf(stderr, "kvm version not supported\n");
1401 s->nr_slots = kvm_check_extension(s, KVM_CAP_NR_MEMSLOTS);
1403 /* If unspecified, use the default value */
1408 s->slots = g_malloc0(s->nr_slots * sizeof(KVMSlot));
1410 for (i = 0; i < s->nr_slots; i++) {
1411 s->slots[i].slot = i;
1414 /* check the vcpu limits */
1415 soft_vcpus_limit = kvm_recommended_vcpus(s);
1416 hard_vcpus_limit = kvm_max_vcpus(s);
1419 if (nc->num > soft_vcpus_limit) {
1421 "Warning: Number of %s cpus requested (%d) exceeds "
1422 "the recommended cpus supported by KVM (%d)\n",
1423 nc->name, nc->num, soft_vcpus_limit);
1425 if (nc->num > hard_vcpus_limit) {
1427 fprintf(stderr, "Number of %s cpus requested (%d) exceeds "
1428 "the maximum cpus supported by KVM (%d)\n",
1429 nc->name, nc->num, hard_vcpus_limit);
1436 kvm_type = qemu_opt_get(qemu_get_machine_opts(), "kvm-type");
1437 if (machine->kvm_type) {
1438 type = machine->kvm_type(kvm_type);
1439 } else if (kvm_type) {
1440 fprintf(stderr, "Invalid argument kvm-type=%s\n", kvm_type);
1445 ret = kvm_ioctl(s, KVM_CREATE_VM, type);
1446 } while (ret == -EINTR);
1449 fprintf(stderr, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret,
1453 fprintf(stderr, "Please add the 'switch_amode' kernel parameter to "
1454 "your host kernel command line\n");
1460 missing_cap = kvm_check_extension_list(s, kvm_required_capabilites);
1463 kvm_check_extension_list(s, kvm_arch_required_capabilities);
1467 fprintf(stderr, "kvm does not support %s\n%s",
1468 missing_cap->name, upgrade_note);
1472 s->coalesced_mmio = kvm_check_extension(s, KVM_CAP_COALESCED_MMIO);
1474 s->broken_set_mem_region = 1;
1475 ret = kvm_check_extension(s, KVM_CAP_JOIN_MEMORY_REGIONS_WORKS);
1477 s->broken_set_mem_region = 0;
1480 #ifdef KVM_CAP_VCPU_EVENTS
1481 s->vcpu_events = kvm_check_extension(s, KVM_CAP_VCPU_EVENTS);
1484 s->robust_singlestep =
1485 kvm_check_extension(s, KVM_CAP_X86_ROBUST_SINGLESTEP);
1487 #ifdef KVM_CAP_DEBUGREGS
1488 s->debugregs = kvm_check_extension(s, KVM_CAP_DEBUGREGS);
1491 #ifdef KVM_CAP_XSAVE
1492 s->xsave = kvm_check_extension(s, KVM_CAP_XSAVE);
1496 s->xcrs = kvm_check_extension(s, KVM_CAP_XCRS);
1499 #ifdef KVM_CAP_PIT_STATE2
1500 s->pit_state2 = kvm_check_extension(s, KVM_CAP_PIT_STATE2);
1503 #ifdef KVM_CAP_IRQ_ROUTING
1504 s->direct_msi = (kvm_check_extension(s, KVM_CAP_SIGNAL_MSI) > 0);
1507 s->intx_set_mask = kvm_check_extension(s, KVM_CAP_PCI_2_3);
1509 s->irq_set_ioctl = KVM_IRQ_LINE;
1510 if (kvm_check_extension(s, KVM_CAP_IRQ_INJECT_STATUS)) {
1511 s->irq_set_ioctl = KVM_IRQ_LINE_STATUS;
1514 #ifdef KVM_CAP_READONLY_MEM
1515 kvm_readonly_mem_allowed =
1516 (kvm_check_extension(s, KVM_CAP_READONLY_MEM) > 0);
1519 ret = kvm_arch_init(s);
1524 ret = kvm_irqchip_create(s);
1530 memory_listener_register(&kvm_memory_listener, &address_space_memory);
1531 memory_listener_register(&kvm_io_listener, &address_space_io);
1533 s->many_ioeventfds = kvm_check_many_ioeventfds();
1535 cpu_interrupt_handler = kvm_handle_interrupt;
1552 static void kvm_handle_io(uint16_t port, void *data, int direction, int size,
1556 uint8_t *ptr = data;
1558 for (i = 0; i < count; i++) {
1559 address_space_rw(&address_space_io, port, ptr, size,
1560 direction == KVM_EXIT_IO_OUT);
1565 static int kvm_handle_internal_error(CPUState *cpu, struct kvm_run *run)
1567 fprintf(stderr, "KVM internal error. Suberror: %d\n",
1568 run->internal.suberror);
1570 if (kvm_check_extension(kvm_state, KVM_CAP_INTERNAL_ERROR_DATA)) {
1573 for (i = 0; i < run->internal.ndata; ++i) {
1574 fprintf(stderr, "extra data[%d]: %"PRIx64"\n",
1575 i, (uint64_t)run->internal.data[i]);
1578 if (run->internal.suberror == KVM_INTERNAL_ERROR_EMULATION) {
1579 fprintf(stderr, "emulation failure\n");
1580 if (!kvm_arch_stop_on_emulation_error(cpu)) {
1581 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1582 return EXCP_INTERRUPT;
1585 /* FIXME: Should trigger a qmp message to let management know
1586 * something went wrong.
1591 void kvm_flush_coalesced_mmio_buffer(void)
1593 KVMState *s = kvm_state;
1595 if (s->coalesced_flush_in_progress) {
1599 s->coalesced_flush_in_progress = true;
1601 if (s->coalesced_mmio_ring) {
1602 struct kvm_coalesced_mmio_ring *ring = s->coalesced_mmio_ring;
1603 while (ring->first != ring->last) {
1604 struct kvm_coalesced_mmio *ent;
1606 ent = &ring->coalesced_mmio[ring->first];
1608 cpu_physical_memory_write(ent->phys_addr, ent->data, ent->len);
1610 ring->first = (ring->first + 1) % KVM_COALESCED_MMIO_MAX;
1614 s->coalesced_flush_in_progress = false;
1617 static void do_kvm_cpu_synchronize_state(void *arg)
1619 CPUState *cpu = arg;
1621 if (!cpu->kvm_vcpu_dirty) {
1622 kvm_arch_get_registers(cpu);
1623 cpu->kvm_vcpu_dirty = true;
1627 void kvm_cpu_synchronize_state(CPUState *cpu)
1629 if (!cpu->kvm_vcpu_dirty) {
1630 run_on_cpu(cpu, do_kvm_cpu_synchronize_state, cpu);
1634 void kvm_cpu_synchronize_post_reset(CPUState *cpu)
1636 kvm_arch_put_registers(cpu, KVM_PUT_RESET_STATE);
1637 cpu->kvm_vcpu_dirty = false;
1640 void kvm_cpu_synchronize_post_init(CPUState *cpu)
1642 kvm_arch_put_registers(cpu, KVM_PUT_FULL_STATE);
1643 cpu->kvm_vcpu_dirty = false;
1646 int kvm_cpu_exec(CPUState *cpu)
1648 struct kvm_run *run = cpu->kvm_run;
1651 DPRINTF("kvm_cpu_exec()\n");
1653 if (kvm_arch_process_async_events(cpu)) {
1654 cpu->exit_request = 0;
1659 if (cpu->kvm_vcpu_dirty) {
1660 kvm_arch_put_registers(cpu, KVM_PUT_RUNTIME_STATE);
1661 cpu->kvm_vcpu_dirty = false;
1664 kvm_arch_pre_run(cpu, run);
1665 if (cpu->exit_request) {
1666 DPRINTF("interrupt exit requested\n");
1668 * KVM requires us to reenter the kernel after IO exits to complete
1669 * instruction emulation. This self-signal will ensure that we
1672 qemu_cpu_kick_self();
1674 qemu_mutex_unlock_iothread();
1676 run_ret = kvm_vcpu_ioctl(cpu, KVM_RUN, 0);
1678 qemu_mutex_lock_iothread();
1679 kvm_arch_post_run(cpu, run);
1682 if (run_ret == -EINTR || run_ret == -EAGAIN) {
1683 DPRINTF("io window exit\n");
1684 ret = EXCP_INTERRUPT;
1687 fprintf(stderr, "error: kvm run failed %s\n",
1688 strerror(-run_ret));
1692 trace_kvm_run_exit(cpu->cpu_index, run->exit_reason);
1693 switch (run->exit_reason) {
1695 DPRINTF("handle_io\n");
1696 kvm_handle_io(run->io.port,
1697 (uint8_t *)run + run->io.data_offset,
1704 DPRINTF("handle_mmio\n");
1705 cpu_physical_memory_rw(run->mmio.phys_addr,
1708 run->mmio.is_write);
1711 case KVM_EXIT_IRQ_WINDOW_OPEN:
1712 DPRINTF("irq_window_open\n");
1713 ret = EXCP_INTERRUPT;
1715 case KVM_EXIT_SHUTDOWN:
1716 DPRINTF("shutdown\n");
1717 qemu_system_reset_request();
1718 ret = EXCP_INTERRUPT;
1720 case KVM_EXIT_UNKNOWN:
1721 fprintf(stderr, "KVM: unknown exit, hardware reason %" PRIx64 "\n",
1722 (uint64_t)run->hw.hardware_exit_reason);
1725 case KVM_EXIT_INTERNAL_ERROR:
1726 ret = kvm_handle_internal_error(cpu, run);
1729 DPRINTF("kvm_arch_handle_exit\n");
1730 ret = kvm_arch_handle_exit(cpu, run);
1736 cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_CODE);
1737 vm_stop(RUN_STATE_INTERNAL_ERROR);
1740 cpu->exit_request = 0;
1744 int kvm_ioctl(KVMState *s, int type, ...)
1751 arg = va_arg(ap, void *);
1754 trace_kvm_ioctl(type, arg);
1755 ret = ioctl(s->fd, type, arg);
1762 int kvm_vm_ioctl(KVMState *s, int type, ...)
1769 arg = va_arg(ap, void *);
1772 trace_kvm_vm_ioctl(type, arg);
1773 ret = ioctl(s->vmfd, type, arg);
1780 int kvm_vcpu_ioctl(CPUState *cpu, int type, ...)
1787 arg = va_arg(ap, void *);
1790 trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg);
1791 ret = ioctl(cpu->kvm_fd, type, arg);
1798 int kvm_device_ioctl(int fd, int type, ...)
1805 arg = va_arg(ap, void *);
1808 trace_kvm_device_ioctl(fd, type, arg);
1809 ret = ioctl(fd, type, arg);
1816 int kvm_has_sync_mmu(void)
1818 return kvm_check_extension(kvm_state, KVM_CAP_SYNC_MMU);
1821 int kvm_has_vcpu_events(void)
1823 return kvm_state->vcpu_events;
1826 int kvm_has_robust_singlestep(void)
1828 return kvm_state->robust_singlestep;
1831 int kvm_has_debugregs(void)
1833 return kvm_state->debugregs;
1836 int kvm_has_xsave(void)
1838 return kvm_state->xsave;
1841 int kvm_has_xcrs(void)
1843 return kvm_state->xcrs;
1846 int kvm_has_pit_state2(void)
1848 return kvm_state->pit_state2;
1851 int kvm_has_many_ioeventfds(void)
1853 if (!kvm_enabled()) {
1856 return kvm_state->many_ioeventfds;
1859 int kvm_has_gsi_routing(void)
1861 #ifdef KVM_CAP_IRQ_ROUTING
1862 return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING);
1868 int kvm_has_intx_set_mask(void)
1870 return kvm_state->intx_set_mask;
1873 void kvm_setup_guest_memory(void *start, size_t size)
1875 #ifdef CONFIG_VALGRIND_H
1876 VALGRIND_MAKE_MEM_DEFINED(start, size);
1878 if (!kvm_has_sync_mmu()) {
1879 int ret = qemu_madvise(start, size, QEMU_MADV_DONTFORK);
1882 perror("qemu_madvise");
1884 "Need MADV_DONTFORK in absence of synchronous KVM MMU\n");
1890 #ifdef KVM_CAP_SET_GUEST_DEBUG
1891 struct kvm_sw_breakpoint *kvm_find_sw_breakpoint(CPUState *cpu,
1894 struct kvm_sw_breakpoint *bp;
1896 QTAILQ_FOREACH(bp, &cpu->kvm_state->kvm_sw_breakpoints, entry) {
1904 int kvm_sw_breakpoints_active(CPUState *cpu)
1906 return !QTAILQ_EMPTY(&cpu->kvm_state->kvm_sw_breakpoints);
1909 struct kvm_set_guest_debug_data {
1910 struct kvm_guest_debug dbg;
1915 static void kvm_invoke_set_guest_debug(void *data)
1917 struct kvm_set_guest_debug_data *dbg_data = data;
1919 dbg_data->err = kvm_vcpu_ioctl(dbg_data->cpu, KVM_SET_GUEST_DEBUG,
1923 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
1925 struct kvm_set_guest_debug_data data;
1927 data.dbg.control = reinject_trap;
1929 if (cpu->singlestep_enabled) {
1930 data.dbg.control |= KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_SINGLESTEP;
1932 kvm_arch_update_guest_debug(cpu, &data.dbg);
1935 run_on_cpu(cpu, kvm_invoke_set_guest_debug, &data);
1939 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
1940 target_ulong len, int type)
1942 struct kvm_sw_breakpoint *bp;
1945 if (type == GDB_BREAKPOINT_SW) {
1946 bp = kvm_find_sw_breakpoint(cpu, addr);
1952 bp = g_malloc(sizeof(struct kvm_sw_breakpoint));
1959 err = kvm_arch_insert_sw_breakpoint(cpu, bp);
1965 QTAILQ_INSERT_HEAD(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
1967 err = kvm_arch_insert_hw_breakpoint(addr, len, type);
1974 err = kvm_update_guest_debug(cpu, 0);
1982 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
1983 target_ulong len, int type)
1985 struct kvm_sw_breakpoint *bp;
1988 if (type == GDB_BREAKPOINT_SW) {
1989 bp = kvm_find_sw_breakpoint(cpu, addr);
1994 if (bp->use_count > 1) {
1999 err = kvm_arch_remove_sw_breakpoint(cpu, bp);
2004 QTAILQ_REMOVE(&cpu->kvm_state->kvm_sw_breakpoints, bp, entry);
2007 err = kvm_arch_remove_hw_breakpoint(addr, len, type);
2014 err = kvm_update_guest_debug(cpu, 0);
2022 void kvm_remove_all_breakpoints(CPUState *cpu)
2024 struct kvm_sw_breakpoint *bp, *next;
2025 KVMState *s = cpu->kvm_state;
2027 QTAILQ_FOREACH_SAFE(bp, &s->kvm_sw_breakpoints, entry, next) {
2028 if (kvm_arch_remove_sw_breakpoint(cpu, bp) != 0) {
2029 /* Try harder to find a CPU that currently sees the breakpoint. */
2031 if (kvm_arch_remove_sw_breakpoint(cpu, bp) == 0) {
2036 QTAILQ_REMOVE(&s->kvm_sw_breakpoints, bp, entry);
2039 kvm_arch_remove_all_hw_breakpoints();
2042 kvm_update_guest_debug(cpu, 0);
2046 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2048 int kvm_update_guest_debug(CPUState *cpu, unsigned long reinject_trap)
2053 int kvm_insert_breakpoint(CPUState *cpu, target_ulong addr,
2054 target_ulong len, int type)
2059 int kvm_remove_breakpoint(CPUState *cpu, target_ulong addr,
2060 target_ulong len, int type)
2065 void kvm_remove_all_breakpoints(CPUState *cpu)
2068 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2070 int kvm_set_signal_mask(CPUState *cpu, const sigset_t *sigset)
2072 struct kvm_signal_mask *sigmask;
2076 return kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, NULL);
2079 sigmask = g_malloc(sizeof(*sigmask) + sizeof(*sigset));
2082 memcpy(sigmask->sigset, sigset, sizeof(*sigset));
2083 r = kvm_vcpu_ioctl(cpu, KVM_SET_SIGNAL_MASK, sigmask);
2088 int kvm_on_sigbus_vcpu(CPUState *cpu, int code, void *addr)
2090 return kvm_arch_on_sigbus_vcpu(cpu, code, addr);
2093 int kvm_on_sigbus(int code, void *addr)
2095 return kvm_arch_on_sigbus(code, addr);
2098 int kvm_create_device(KVMState *s, uint64_t type, bool test)
2101 struct kvm_create_device create_dev;
2103 create_dev.type = type;
2105 create_dev.flags = test ? KVM_CREATE_DEVICE_TEST : 0;
2107 if (!kvm_check_extension(s, KVM_CAP_DEVICE_CTRL)) {
2111 ret = kvm_vm_ioctl(s, KVM_CREATE_DEVICE, &create_dev);
2116 return test ? 0 : create_dev.fd;